Entertainment Computing – ICEC 2022: 21st IFIP TC 14 International Conference, ICEC 2022, Bremen, Germany, November 1–3, 2022, Proceedings (Lecture Notes in Computer Science) 3031202112, 9783031202117

This book constitutes the refereed proceedings of the 21sth IFIP TC 14 International Conference on Entertainment Computi

151 43 49MB

English Pages 472 [456] Year 2022

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Preface
Organization
Contents
Game Experience
AstraVerse: Establishing a Culturally Sensitive Framework for Integrating Elements from Mythological Backgrounds
1 Introduction
2 Related Work
2.1 Cultural Affordances in Games
2.2 Theology and Gaming
2.3 Game Design Frameworks and Taxonomies
3 AstraVerse Framework
3.1 Mythological Derivation
3.2 Generative and Evaluative Steps
4 Applying the Framework on Hindu Mythology
4.1 Mythological Derivation [Step 1]: Establishing a Taxonomy on Mythological Hindu Characters and Mythical Objects
4.2 Generative Step [Step 2]: Facilitating the Creation of Characters from the Established Hindu Taxonomy
4.3 Constructed Outcomes from Generative Step
4.4 Creative Evaluation [Step 3(a)]: Evaluating the Creative Dimension of Workshop Outcomes
4.5 Cultural Evaluation [Step 3(b)]: Evaluating the Cultural Validity of Workshop Outcomes
5 Results and Discussion
6 Future Work and Conclusion
References
Comfortably Numb? Violent Video Games and Their Effects on Aggression, Mood, and Pain-Related Responses
1 Introduction
1.1 Violent Video Games and Aggression Desensitization
1.2 Violent Video Games and Pain Desensitization
1.3 Hypotheses
2 Study 1
2.1 Methods
2.2 Results
2.3 Discussion
3 Study 2
3.1 Methods
3.2 Results
3.3 Discussion
4 Study 3
4.1 Methods
4.2 Results
4.3 Discussion
5 General Discussion
References
Discovering the Motivational Constitution of `Playing Games for Fun'
1 Introduction
2 Methods
3 Results
4 Concluding Statements
References
Towards an Understanding of How Players Make Meaning from Post-Play Process Visualizations
1 Introduction
2 Related Work
3 Methodology
3.1 Interview Protocol
3.2 Data Analysis
4 Results
4.1 Sense-Making Methods
5 Discussion and Implications
6 Limitations
7 Conclusion
References
Entertainment Tools and Methods
Plot Composition by Mapping Situation Calculus Schemas into Petri Net Representation
1 Introduction
2 Related Work
3 From Situation Calculus to Petri Net Models
3.1 The Basic Situation Calculus Model
3.2 Deriving a Petri Net from a Situation Calculus Model
4 Interactive Storytelling Application
4.1 System Architecture
4.2 Interactive Dramatization
5 Concluding Remarks
References
Upward Influence Tactics: Playful Virtual Reality Approach for Analysing Human Multi-robot Interaction
1 Introduction
2 Related Work
3 Approach
3.1 Robot Avatar
3.2 Virtual Environment and Interaction Modes
3.3 Task of the Game
3.4 Robot Behavior
4 Evaluation
4.1 Method
4.2 Findings
5 Discussion
6 Conclusion
References
OptimizingMARL: Developing Cooperative Game Environments Based on Multi-agent Reinforcement Learning
1 Introduction
2 Related Work
3 The Framework OptimizingMARL
3.1 Model the Multi-agent Learning Environment
3.2 Design Rewards and Distributing Knowledge
3.3 Create the Learning Environment
4 Experiments
4.1 Dungeon Escape Environment
4.2 Color Balls Environment
4.3 Wild World Environment
5 Conclusion
References
Game Engine Comparative Anatomy
1 Introduction
2 Related Work
3 Approach
3.1 Overview
3.2 Engine Selection
3.3 Engine Compilation and Analysis
4 Godot's Call Graph
5 Urho3D's Call Graph
6 Comparing Godot and Urho3D
7 Discussion
8 Conclusion
References
Design of an Extended Reality Collaboration Architecture for Mixed Immersive and Multi-surface Interaction
1 Introduction
2 Related Work
3 Materials and Methods
3.1 2D Visualization
3.2 AR Visualization
4 Future Work and Limitations
5 Conclusion
References
Player Behavior and Analysis
Mental Wear and Tear: An Exploratory Study on Mental Fatigue in Video Games Using the Example of League of Legends
1 Introduction
2 Related Work
3 Study
3.1 Measures
3.2 Participants
4 Results
5 Discussion
6 Limitations and Future Work
7 Conclusion
References
Difficulty Pacing Impact on Player Motivation
1 Introduction
2 Difficulty and Motivation
3 Difficulty Curves
4 Experimentation
4.1 Hypotheses
4.2 Methodology
5 Results
6 Discussion
7 Conclusion
References
Toward Dynamic Difficulty Adjustment with Audio Cues by Gaussian Process Regression in a First-Person Shooter
1 Introduction
2 Related Work
2.1 Audio Cues
2.2 GPR
3 Methodology
3.1 Difficulty Recommendation
3.2 Difficulty Setting with GPR
4 Experiment and Result
5 Conclusions and Future Work
References
Playstyles in Tetris: Beyond Player Skill, Score, and Competition
1 Introduction
2 Background
3 Tetris and Our Variations
4 Method
5 Results
6 Discussion and Limitations
7 Conclusion
References
A Matter of Closeness: Player-Avatar Relationships as Degree of Including Avatars in the Self
1 Introduction
2 Theoretical Approaches to Player-Avatar Relationships
2.1 Status Quo
2.2 The Inclusion-of-Other-In-The-Self Principle
2.3 PAR as Inclusion of the Avatar in the Self
3 Method
4 Results
4.1 A Distance-Closeness Continuum
4.2 In-Depth Findings: Five Types of PARs and Related Experiences
5 Discussion
References
Serious Gameplay
Through Troubled Waters: A Narrative Game for Anger Regulation
1 Introduction
2 Game Design
2.1 Original Game: When Life Gives You Lemons (WLGYL)
2.2 Modified Game: Through Troubled Waters (TTW)
3 Method
3.1 Measurements
3.2 Sampling and Experiment Procedure
3.3 Data Analysis
4 Results
4.1 Quantitative Results
4.2 Qualitative Results
5 Discussion
6 Conclusion
References
Incorporating the Theory of Attention in Applied Game Design
1 Introduction
2 Related Work
2.1 Areas of Involvement
2.2 Balance and Rhythm
2.3 Attention in Applied Games
3 Implementation
3.1 Case 1: Virtual Reality (VR) Person-Centred Care (PCC)
3.2 Case 2: When Life Gives You Lemons (WLGYL)
3.3 The Lens of Attention
3.4 Design Discussions
4 Discussion
5 Conclusion
References
Exergames in the GAME2AWE Platform with Dynamic Difficulty Adjustment
1 Introduction
2 GAME2AWE Platform
2.1 Scope
2.2 Methodology
2.3 Example Games
3 Dynamic Difficulty Adjustment
3.1 Background
3.2 GAME2AWE Approach
3.3 Building the Fun Prediction Model
4 Conclusions
References
Art and Entertainment
Improvement of Deep Learning Technology to Create 3D Model of Fluid Art
1 Introduction
2 Sound of Ikebana
3 3D Reconstruction of Sound of Ikebana
3.1 Phase only Correlation Method
3.2 GANs
3.3 Improvement of 3D Modeling for Sound of Ikebana
3.4 Results
4 Conclusion
References
Method for Music Game Control Using Myoelectric Sensors
1 Introduction
2 Music Game Control Optimization System
2.1 System Implementation
2.2 Sensing Devices for Measuring EMG
2.3 Evaluation of Muscle Fatigue
2.4 Music Game
3 Evaluation of the Proposed Method
3.1 Experimental Method
3.2 Experimental Results
3.3 Discussion
4 Conclusion and Future Work
References
Creation of Fluid Art “Sound of Ikebana” Under Microgravity Using Parabolic Flight
1 Introduction
2 Related Works
3 Fluid Art “Sound of Ikebana”
3.1 Fluid Art
3.2 “Sound of Ikebana”
4 Microgravity Generation Method
4.1 Parabolic Flight
4.2 Free-Fall
5 Creation of Fluid Art Under Microgravity
5.1 Basic Methodology
5.2 Sound of Ikebana Creation Using Parabolic Flight
5.3 Sound of Ikebana Under the Parabolic Flight
5.4 Three-Dimensional Materialization of the Sound of Ikebana Under Microgravity
6 Conclusion
References
Design Explorations of Interactive Point Cloud Based Virtual Environments Using Volumetric Capture and Visualisation Techniques
1 Introduction
2 Related Work
3 Development Pipeline
4 Prototype: Tilia x europaea
4.1 Point-Cloud Representation
4.2 Representation of Natural Processes
4.3 Interactions
4.4 Sound
4.5 Locomotion and Area Traversal
5 Evaluation
6 Conclusion and Future Work
References
Game Communities
Predicting Success Factors of Video Game Titles and Companies
1 Introduction
2 Related Work
3 Approach
3.1 Dataset and Processing
3.2 Prediction Outcomes
4 Validation
4.1 Measures
4.2 Procedure
4.3 Participants
5 Results and Discussion
5.1 General Findings
5.2 Survivability
5.3 Player Ratings
5.4 Return on Investment
6 Implications
7 Limitations and Future Work
8 Conclusion
A Appendix
A.1 Survivability
A.2 Player Rating
A.3 Sharpe Ratio (ROI)
References
The Importance of Dashboard Elements During Esports Matches to Players, Passive-Viewers and Spectator-Players
1 Introduction
2 Related Work
3 Design
4 Method
4.1 Participants
4.2 Procedure
4.3 Data Analysis
5 Results
5.1 External Scoreboard
5.2 Internal Scoreboard
5.3 Direct Opponent
5.4 Last Teamfight
5.5 Webcams
5.6 Qualitative Feedback
6 Discussion
7 Limitations and Future Work
8 Conclusion
References
.26em plus .1em minus .1emA Reusable Methodology for Player Clustering Using Wasserstein Autoencoders
1 Introduction
2 Background
2.1 Recurrent Neural Networks
2.2 Autoencoders
2.3 Adversarial Training and Generative Adversarial Networks
2.4 Wasserstein Distance
2.5 Clustering Metrics
2.6 Archetypal Analysis
3 Related Work
4 Method
4.1 Game Identification and Understanding: My Singing Monsters
4.2 Data Collection and Preprocessing
4.3 Model Definition and Construction
5 Translation into Actionable Interpretations: Clustering Analysis
5.1 Player Visualizations
5.2 Representative Analysis
6 Discussion
References
Towards a Community-Based Ranking System of Overwatch Players
1 Introduction
2 Background
3 Overwatch and Player Rankings
4 Method
4.1 Data
4.2 Survey
4.3 Rating System
5 Use Cases
6 Discussion
7 Conclusions and Future Work
References
Understanding Stakeholders’ Perspectives Towards Serious Games for Vocational Training for People with Intellectual Disabilities in Macau
1 Introduction
2 Related Work
3 Understanding Stakeholders’ Perspectives
3.1 Questionnaires
3.2 Interviews
3.3 Discussion
4 Conclusion
Appendix A
References
Workshop: Digital Arts and Health
SleepHill: Designing an Incrementally Bouncing Pillow as a Comfortable Wake-Up Approach
1 Introduction
2 SleepHill
2.1 System Design
2.2 Implementation
3 The Study
3.1 Participants
3.2 Apparatus and Setup
3.3 Procedure
3.4 Data Collection and Analysis
4 Findings
4.1 Quantitative Findings
4.2 Qualitative Findings
5 Discussion, Conclusions, and Future Work
5.1 Design Implications
5.2 Limitation and Future Work
References
Digital-Pen: An Interactive System for Correcting Writing Posture of Primary School Students
1 Introduction
2 Related Work
2.1 Research on Smart Pen
2.2 Research on the Hidden Health Problems in the Writing Process of Primary School Students
2.3 The Data and Explanation of the Correct Writing Posture of Primary School Students
3 Design Process of the Digital-Pen Interactive System
3.1 Design Exploration with Experts
3.2 Design Goals
4 System Design
4.1 Application Design of the Software System
4.2 Design of Smart Hardware Pen
4.3 Communication and Data Processing of Hardware and Software
5 Preliminary User Research
5.1 Procedure
5.2 Results
6 Current Limitations and Future Work
7 Conclusion
References
The Study on Digital Art Generation of Health Data for the Elderly
1 Introduction
2 Related Work
3 System Design
3.1 The Analysis of Health Data Generates Digital Art
3.2 The Description of Visual System Design
3.3 The Construction of Algorithm Model and Framework
3.4 Final Visual Presentation
4 Analysis
4.1 Study Methods
4.2 Study Results
5 Result
6 Conclusion
References
Effects of Color Tone of Dynamic Digital Art on Emotion Arousal
1 Introduction
2 Related Work
3 Study Design
3.1 Stimuli
3.2 Participants
3.3 Experiment Design
4 Discussion and Future Work
5 Conclusion
References
Personalized Synchronous Running Music Remix Procedure for Novice Runners
1 Introduction
2 Related Work
2.1 Music, Health, and Sports
2.2 Intervention of Music on Sports Performance
3 Proposed Method
3.1 Trial Run
3.2 Data Analytic and Goal Establishment
3.3 Music Retrieval and Pre-processing
3.4 Mixing
4 Experiment
4.1 Experiment Design
4.2 Procedure
5 Results
5.1 Quantitative Analysis
5.2 Qualitative Analysis
6 Discussion
Appendix 1. Power Words List
References
Design for Connecting People Through Digital Artworks with Personal Information
1 Introduction
2 Related Work
2.1 Social Connectedness and Health
2.2 Work for Social Connectedness Enhancement Remotely
3 Design
3.1 Proposed Design
3.2 Design Considerations
3.3 Design Components
4 Pilot Study
4.1 Experiment Setup and Place
4.2 Participants, Procedure and Measurements
4.3 Preliminary User Feedback
5 Discussion
6 Conclusion and Future Work
References
Research on the Design of the Pre-hospital Emergency System Based on Kano-QFD
1 Introduction
2 Pre-hospital Emergency System
3 Explanation of the Kano-QFD Method
4 Research Process of Pre-hospital Emergency System Design Based on Kano-QFD
4.1 User Requirement Elicitation
4.2 Analysis of Interactive Requirements for the Pre-hospital Emergency System
4.3 QFD-Based Functional Transformation and Sorting
5 Design of the Pre-hospital Emergency System
5.1 Manual for and Structure of the Pre-hospital Emergency System
5.2 Design of the Scheme
6 Conclusion
Appendix
References
Student Game Competition
The Public Transport Stress Test (PTST) at Bogotá: A Perceived Reality and Handling Anxious Situations with a Cool Mind Through Preparation
1 Motivation
1.1 Problem Statement
1.2 Goals
2 Game Experience
2.1 Abstraction
2.2 Game Structure
2.3 Aesthetics
2.4 Minigames and Mechanics
2.5 Additional Content
3 Proposed Solution
3.1 Results
References
Corruptio: An Interactive Insight on City Management
1 Introduction
1.1 Development Process
2 Game Experience
2.1 Base City
2.2 Match System
2.3 Budget System
2.4 Adaptative City System
2.5 Notifications System
2.6 Decision-Making System
2.7 Random Events System
3 Results
References
Workshops and Tutorials
Workshop on Social and Ethical Issues in Entertainment Computing
1 Introduction
2 Background
2.1 Foundations
2.2 Ethics While Playing Computer Games
2.3 Social Issues While Playing Computer Games
2.4 Legal Issues
2.5 Digital Game-Based Learning
2.6 Monetization
2.7 Diversity
3 Conclusions
References
Digital Arts and Health
1 Arts and Health
2 Digital Arts and Health
3 Opportunities and Challenges
4 Workshop Goals
5 Summary
References
Current Opportunities and Challenges of Digital Game-Based Learning
1 Introduction
2 Life Cycle Model for Digital Learning Games
3 Opportunities and Challenges
3.1 Ideation
3.2 Design
3.3 Implementation
3.4 Evaluation
3.5 Operation
3.6 Maintenance
4 Conclusion
References
Author Index
Recommend Papers

Entertainment Computing – ICEC 2022: 21st IFIP TC 14 International Conference, ICEC 2022, Bremen, Germany, November 1–3, 2022, Proceedings (Lecture Notes in Computer Science)
 3031202112, 9783031202117

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

LNCS 13477

Barbara Göbl Erik van der Spek Jannicke Baalsrud Hauge Rod McCall (Eds.)

Entertainment Computing – ICEC 2022 21st IFIP TC 14 International Conference, ICEC 2022 Bremen, Germany, November 1–3, 2022 Proceedings

Lecture Notes in Computer Science Founding Editors Gerhard Goos Karlsruhe Institute of Technology, Karlsruhe, Germany Juris Hartmanis Cornell University, Ithaca, NY, USA

Editorial Board Members Elisa Bertino Purdue University, West Lafayette, IN, USA Wen Gao Peking University, Beijing, China Bernhard Steffen TU Dortmund University, Dortmund, Germany Moti Yung Columbia University, New York, NY, USA

13477

More information about this series at https://link.springer.com/bookseries/558

Barbara Göbl · Erik van der Spek · Jannicke Baalsrud Hauge · Rod McCall (Eds.)

Entertainment Computing – ICEC 2022 21st IFIP TC 14 International Conference, ICEC 2022 Bremen, Germany, November 1–3, 2022 Proceedings

Editors Barbara Göbl University of Vienna Vienna, Austria Jannicke Baalsrud Hauge Royal Institute of Technology Södertälje, Sweden BIBA – Bremer Institut für Produktion und Logistik GmbH Bremen, Germany

Erik van der Spek Eindhoven University of Technology Eindhoven, The Netherlands Rod McCall Luxembourg Institute of Science and Technology Esch-Sur-Alzette, Luxembourg

ISSN 0302-9743 ISSN 1611-3349 (electronic) Lecture Notes in Computer Science ISBN 978-3-031-20211-7 ISBN 978-3-031-20212-4 (eBook) https://doi.org/10.1007/978-3-031-20212-4 © IFIP International Federation for Information Processing 2022 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

We are proud to present the conference proceedings of the 21st edition of the IFIP International Conference on Entertainment Computing (ICEC 2022) in this edited LNCS volume. The conference was hosted by “BIBA - Bremer Institut für Produktion und Logistik an der Universität Bremen” in Bremen, Germany, during November 1–3, 2022. As the longest lasting and prime scientific conference in the area of Entertainment Computing, ICEC brings together researchers and practitioners with diverse backgrounds in order to connect, share, and discuss both recent and potential future developments in this field. Considering the broad range of topics represented in our three keynotes, 35 papers, four workshops, and one tutorial, ICEC 2022 served as a lively forum for multidisciplinary exchange to advance our understanding of Entertainment Computing and all related areas. Overall, we received a total of 72 submissions from authors in Europe, Asia, North and South America, and Oceania. The works collected in this volume discuss latest findings in the fields of Entertainment Tools and Methods, Serious Gameplay, Game Communities, Player Behaviour and Analysis, Game Experience, and Art and Entertainment. All papers underwent double blind peer review with an average of 3.2 reviews per paper. This resulted in 13 accepted full papers, 13 work-in-progress papers, two student competition papers, and three workshop papers. Additionally, a special track presented seven peer-reviewed papers in the fields of Digital Arts and Health, discussed in a workshop hosted by IFIP Working Group (WG) 14.7. Three further workshops were organized on the topics of Social and Ethical Issues in Entertainment Computing (WG 14.5), Interactive Immersive Entertainment (WG 14.6), and Current Opportunities and Challenges of Digital Game-based Learning (WG 14.8). We would like to express our gratitude to everyone who supported us in hosting this year’s conference and ensuring the high quality of the presented proceedings. We thank all members of the Program Committee, composed of 91 experts from 29 different countries, for their hard work. We would also like to thank all the Organizing Committee members who contributed their valuable time and insights. Many thanks also go to the International Federation for Information Processing (IFIP), the ICEC Steering Committee, and our local organizers. Without all this support, this conference would not have been possible. September 2022

Barbara Göbl Erik van der Spek Jannicke Baalsrud Hauge Rod McCall

Organization

General Chairs Jannicke Baalsrud Hauge Rod McCall

KTH Royal Institute of Technology, Sweden, and BIBA Bremen, Germany Luxembourg Institute of Science and Technology, Luxembourg

Program Committee Chairs Barbara Göbl Erik van der Spek

University of Vienna, Austria Eindhoven University of Technology, The Netherlands

Work in Progress Chairs Jorge C. S. Cardoso Dimitry Alexandrovsky

University of Coimbra, Portugal University of Bremen, Germany

Tutorials Chair Esteban Clua

Federal Fluminense University, Brazil

Workshop Chairs Heinrich Söbke Paula Alexandra Silva

Bauhaus-Universität Weimar, Germany University of Coimbra, Portugal

Interactive Entertainment/Experiential Works Chair Moritz Quant

BIBA Bremen, Germany

Student Competition Chairs Thiago Porcino Jaime Garcia

Dalhousie University, Canada University of Technology Sydney, Australia

viii

Organization

Doctoral Consortium Chairs Helmut Hlavacs Simone Kriglstein

University of Vienna, Austria Masaryk University, Czech Republic, and AIT Austrian Institute of Technology GmbH, Austria

Publicity Chair Aleksandra Himstedt

BIBA Bremen, Germany

Local Organization Sundus Fatima Moritz Quandt Heiko Duin Jakob Baalsrud Hauge

BIBA Bremen, Germany BIBA Bremen, Germany BIBA Bremen, Germany Hochschule Bremen, Germany

Steering Committee Licinio Roque Jannicke Baalsrud Hauge Esteban Clua Jorge C. S. Cardoso Rainer Malaka Erik van der Spek

University of Coimbra, Portugal KTH Royal Institute of Technology, Sweden, and BIBA Bremen, Germany Federal Fluminense University, Brazil University of Coimbra, Portugal University of Bremen, Germany Eindhoven University of Technology, Netherlands

Program Committee Matthew Barr Nicole Basaraba Regina Bernhaupt Ioannis Bikas Fernando Birra Nicholas Bowman Joseph Alexander Brown Carlos Caires Elin Carstensdottir Joao Cordeiro Nuno Correia Drew Davidson

University of Glasgow, UK Maastricht University, The Netherlands Eindhoven University of Technology, The Netherlands University of Bremen, Germany Universidade NOVA de Lisboa, Portugal Texas Tech University, USA Brock University, Canada University of Saint Joseph, Macao University of California, Santa Cruz, USA Portuguese Catholic University, Portugal Universidade NOVA de Lisboa, Portugal Carnegie Mellon University, USA

Organization

Mara Dionisio Ralf Doerner Heiko Duin Kai Erenli Gerald Estadieu Bruno Feijo Pablo Figueroa Mateus Finco Mikael Fridenfalk Florian Gnadlinger

Stefan Goebel Pedro Gonzalez Calero Noriko Hanakawa Robin Horst Jun Hu Yuqi Hu Hiroyuki Iida Naoya Isoyama Jinyuan Jia Hao Jiang Philipp Jordan Fares Kayali Mohd Nor Akmal Khalid Chris Kiefer Christoph Klimmt Kei Kobayashi Troy Kohwalter Sara Kunz Mei-Kei Lai Danielle Langlois Michael Lankes Jingya Li

ix

Madeira Interactive Technologies Institute, Portugal RheinMain University of Applied Sciences, Germany BIBA Bremen, Germany UAS BFI Vienna, Austria University of Saint Joseph, Macao Pontifical Catholic University of Rio de Janeiro, Brazil Universidad de los Andes, Colombia Universidade Federal do Rio Grande do Sul, Brazil Uppsala University, Sweden University of Applied Sciences HTW Berlin, Germany, and Masaryk University, Czech Republic TU Darmstadt, Germany Universidad Politécnica de Madrid, Spain Hannan University, Japan RheinMain University of Applied Sciences, Germany Eindhoven University of Technology, The Netherlands University of Nottingham Ningbo China, China JAIST, Japan Nara Institute of Science and Technology, Japan Tongji University, China Zhejiang University, Ningbo, China University of Hawaii at Manoa, USA University of Vienna, Austria Japan Advanced Institute of Science and Technology, Japan University of Sussex, UK Hanover University of Music, Drama, and Media, Germany Nagoya City University, Japan Universidade Federal Fluminense, Brazil Portuguese Catholic University, Portugal Macao Polytechnic University, Macao Masaryk University, Czech Republic University of Applied Sciences Upper Austria, Hagenberg, Austria Beijing Jiaotong University, China

x

Organization

Qingde Li Sheng Li Bingjian Liu Penousal Machado Hung Mai Cong Panos Markopoulos Filipa Martins de Abreu Andre Miede Michela Mortara Wolfgang Mueller Satoshi Nakamura Ryohei Nakatsu Andrés Adolfo Navarro Newball Carla Patrão Andre Perrotta Johannes Pfau Maria Popescu Kjetil Raaen Theresa-Marie Rhyne Teresa Romeo Licinio Roque Che Mat Ruzinoor Anthony Savidis Andreas Scalas Mariana Seiça Nikitas Sgouros Yang Shen Dirk Snyman Edirlei Soares de Lima Ioana Andreea Stefan Matthias Steinböck Elif Surer Cristina Sylla Laszlo Szecsi Daniel Thalmann Joseph Timoney Mai Xuan Trang Kai Tuuri

University of Hull, UK Peking University, China University of Nottingham Ningbo China, China University of Coimbra, Portugal Kyoto University, Japan Eindhoven University of Technology, The Netherlands CITAR, Portugal Hochschule für Technik und Wirtschaft des Saarlandes, Germany CNR-IMATI, Italy University of Education Weingarten, Germany Meiji University, Japan Kyoto University, Japan Pontificia Universidad Javeriana, Cali, Colombia Instituto Politécnico de Coimbra, Portugal CITAR, Portugal Universität Bremen, Germany Carol I National Defence University, Romania Høyskolen Kristiania, Norway Independent Consultant, USA Universidade NOVA de Lisboa, Portugal University of Coimbra, Portugal Universiti Putra Malaysia, Malaysia University of Crete and ICS-FORTH, Greece CNR-IMATI, Italy University of Coimbra, Portugal University of Piraeus, Greece Beijing Normal University, China North-West University, South Africa Universidade Europeia, Portugal Advanced Technology Systems, Romania University of Vienna, Austria Middle East Technical University, Turkey University of Minho, Portugal Budapest University of Technology and Economics, Hungary Ecole Polytechnique Fédérale de Lausanne, Switzerland Maynooth University, Ireland Phenikaa University, Vietnam University of Jyväskylä, Finland

Organization

Guenter Wallner Xinwei Wang Robert Wendrich Jiang Wu Mengru Xue Xiaosong Yang Cheng Yao Nelson Zagalo Leijing Zhou

Johannes Kepler University Linz, Austria University of Nottingham Ningbo China, China University of Twente, The Netherlands University of Nottingham Ningbo China, China Zhejiang University, Ningbo, China Bournemouth University, UK Zhejiang University, China University of Aveiro, Portugal Zhejiang University, China

xi

Contents

Game Experience AstraVerse: Establishing a Culturally Sensitive Framework for Integrating Elements from Mythological Backgrounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sai Siddartha Maram, Johannes Pfau, Reza Habibi, and Magy Seif El-Nasr Comfortably Numb? Violent Video Games and Their Effects on Aggression, Mood, and Pain-Related Responses . . . . . . . . . . . . . . . . . . . . . . . . . Gary L. Wagener and André Melzer Discovering the Motivational Constitution of ‘Playing Games for Fun’ . . . . . . . . Kai Tuuri and Jukka Vahlo Towards an Understanding of How Players Make Meaning from Post-Play Process Visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Erica Kleinman, Jennifer Villareale, Murtuza Shergadwala, Zhaoqing Teng, Andy Bryant, Jichen Zhu, and Magy Seif El-Nasr

3

18

39

47

Entertainment Tools and Methods Plot Composition by Mapping Situation Calculus Schemas into Petri Net Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Edirlei Soares de Lima, Antonio L. Furtado, Bruno Feijó, and Marco A. Casanova Upward Influence Tactics: Playful Virtual Reality Approach for Analysing Human Multi-robot Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cornelia Gerdenitsch, Matthias Weinhofer, Jaison Puthenkalam, and Simone Kriglstein OptimizingMARL: Developing Cooperative Game Environments Based on Multi-agent Reinforcement Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thaís Ferreira, Esteban Clua, Troy Costa Kohwalter, and Rodrigo Santos

61

76

89

Game Engine Comparative Anatomy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Gabriel C. Ullmann, Cristiano Politowski, Yann-Gaël Guéhéneuc, and Fabio Petrillo

xiv

Contents

Design of an Extended Reality Collaboration Architecture for Mixed Immersive and Multi-surface Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Thiago Porcino, Seyed Adel Ghaeinian, Juliano Franz, Joseph Malloch, and Derek Reilly Player Behavior and Analysis Mental Wear and Tear: An Exploratory Study on Mental Fatigue in Video Games Using the Example of League of Legends . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Ioannis Bikas, Johannes Pfau, Bastian Dänekas, and Rainer Malaka Difficulty Pacing Impact on Player Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 William Rao Fernandes and Guillaume Levieux Toward Dynamic Difficulty Adjustment with Audio Cues by Gaussian Process Regression in a First-Person Shooter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 Xiaoxu Li, Marcel Wira, and Ruck Thawonmas Playstyles in Tetris: Beyond Player Skill, Score, and Competition . . . . . . . . . . . . 162 Stéphanie Mader and Eloïse Tassin A Matter of Closeness: Player-Avatar Relationships as Degree of Including Avatars in the Self . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Daniel Possler, Natascha N. Carnol, Christoph Klimmt, Ina Weber-Hoffmann, and Arthur A. Raney Serious Gameplay Through Troubled Waters: A Narrative Game for Anger Regulation . . . . . . . . . . . 185 Jiaqi Li, Sotirios Piliouras, Semma Raadschelders, Vivian Imani Dap, Claudia Alessandra Libbi, and Marcello A. Gómez-Maureira Incorporating the Theory of Attention in Applied Game Design . . . . . . . . . . . . . . 200 Isabelle Kniestedt, Stephan Lukosch, Milan van der Kuil, Iulia Lefter, and Frances Brazier Exergames in the GAME2AWE Platform with Dynamic Difficulty Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 Michail Danousis, Christos Goumopoulos, and Alexandros Fakis

Contents

xv

Art and Entertainment Improvement of Deep Learning Technology to Create 3D Model of Fluid Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 Mai Cong Hung, Mai Xuan Trang, Akihiro Yamada, Naoko Tosa, and Ryohei Nakatsu Method for Music Game Control Using Myoelectric Sensors . . . . . . . . . . . . . . . . 238 Shuo Zhou and Norihisa Segawa Creation of Fluid Art “Sound of Ikebana” Under Microgravity Using Parabolic Flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 Naoko Tosa, Shigetaka Toba, Yunian Pang, Akihiro Yamada, Takashi Suzuki, and Ryohei Nakatsu Design Explorations of Interactive Point Cloud Based Virtual Environments Using Volumetric Capture and Visualisation Techniques . . . . . . . . 256 Maximilian Rubin, Jorge C. S. Cardoso, and Pedro Martins Carvalho Game Communities Predicting Success Factors of Video Game Titles and Companies . . . . . . . . . . . . . 269 Johannes Pfau, Michael Debus, Jesper Juul, Emil Lundedal Hammar, Alessandro Canossa, and Magy Seif El-Nasr The Importance of Dashboard Elements During Esports Matches to Players, Passive-Viewers and Spectator-Players . . . . . . . . . . . . . . . . . . . . . . . . . . 283 Stan J. P. van Kempen, Erik D. van der Spek, and Günter Wallner A Reusable Methodology for Player Clustering Using Wasserstein Autoencoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 Jonathan Tan and Mike Katchabaw Towards a Community-Based Ranking System of Overwatch Players . . . . . . . . . 309 David Staat, Günter Wallner, and Regina Bernhaupt Understanding Stakeholders’ Perspectives Towards Serious Games for Vocational Training for People with Intellectual Disabilities in Macau . . . . . . 320 Choi On Lei and Jorge C. S. Cardoso

xvi

Contents

Workshop: Digital Arts and Health SleepHill: Designing an Incrementally Bouncing Pillow as a Comfortable Wake-Up Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 Wenshu Xun, Pengsong Zhang, Zixuan Liu, Yufei Meng, Yaqi Zheng, Renyao Zou, and Xipei Ren Digital-Pen: An Interactive System for Correcting Writing Posture of Primary School Students . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342 Yousheng Yao, Jiacheng Lou, Guanghui Huang, Xuesong Li, and Yingrui Li The Study on Digital Art Generation of Health Data for the Elderly . . . . . . . . . . . 354 Shuyao Li, Yahi Shuai, Ran Wan, Jianxin Jin, Shuqi Wang, Zepeng Yu, Qianqian Hu, Feifei Liu, and Yijia Wang Effects of Color Tone of Dynamic Digital Art on Emotion Arousal . . . . . . . . . . . 363 Qiurui Wang, Zhenyu Liu, and Jun Hu Personalized Synchronous Running Music Remix Procedure for Novice Runners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372 Nan Zhuang, Shitong Weng, Song Bao, Xinyi Li, Jingru Huang, and Pinhao Wang Design for Connecting People Through Digital Artworks with Personal Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 Rui Wang and Jun Hu Research on the Design of the Pre-hospital Emergency System Based on Kano-QFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398 Di Lu, Guanghui Huang, Fangtian Ying, and Yu Jiang Student Game Competition The Public Transport Stress Test (PTST) at Bogotá: A Perceived Reality and Handling Anxious Situations with a Cool Mind Through Preparation . . . . . . 415 Juan Pablo Romero, Juan Diego Rueda, José David Tamara, and Pablo Figueroa Corruptio: An Interactive Insight on City Management . . . . . . . . . . . . . . . . . . . . . . 421 Andrés Sánchez, Santiago Gamboa, Nicolas Fajardo, and Pablo Figueroa

Contents

xvii

Workshops and Tutorials Workshop on Social and Ethical Issues in Entertainment Computing . . . . . . . . . . 429 Roderick McCall, Jethro Shell, Celina Kacperski, Stanley Greenstein, Nicola Whitton, and Jo Summers Digital Arts and Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436 Mengru Xue, Cheng Yao, Jun Hu, Yuqi Hu, and Hui Lyu Current Opportunities and Challenges of Digital Game-Based Learning . . . . . . . 443 Jannicke Baalsrud Hauge, Heinrich Söbke, Heiko Duin, Ioana Andreea Stefan, and Barbara Göbl Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451

Game Experience

AstraVerse: Establishing a Culturally Sensitive Framework for Integrating Elements from Mythological Backgrounds Sai Siddartha Maram(B) , Johannes Pfau , Reza Habibi , and Magy Seif El-Nasr University of California, Santa Cruz, CA, USA {samaram,jopfau,rehabibi,mseifeln}@uscc.edu

Abstract. Game designers often refer to sources of inspiration for creating new characters and narratives, with cultural and mythological references being a prominent choice for orientation. This however implicitly entails the importance of preventing the misuse of mythological or religious references to respect affiliated groups and believers. In this paper, we present a three-step framework to augment the ability of designers to create innovative characters using mythology as an inspiration while addressing and avoiding cultural backlash. The efficacy of the framework is verified by implementing the framework on Hindu Mythology. We present qualitative and quantitative findings which indicate that characters designed using the framework report significantly higher appraisal from a Hindu population than controversial portrayals of Hindu deities from actually shipped games.

Keywords: Game design

1

· Character design · Co-design · Culture

Introduction

With trends of game design and development moving towards more agile approaches, it becomes challenging for game designers to constantly produce unique characters, narratives and mechanics. Olesen reports how it is common to hear from video game designers in agile environments that they find it difficult to generate ideas for narratives and inspiration for character design [31]. Considering established literature in the field of design, inspiration is one of the main factors for generating new ideas within game development [9,25,36,46]. According to Hagen [19], one of the most popular sources of inspiration is mythology as to help in designing new characters. Drawing from mythology makes for a frequent occurrence within the gaming industry, but has not had much discussion in academia [12,48]; especially Roman and Greek mythology have commonly been utilized to create characters [16]. Successful known titles include gods and characters transplanted from various mythologies [4,30,41,42,45]. Not only does the mere transplantation of gods c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 3–17, 2022. https://doi.org/10.1007/978-3-031-20212-4_1

4

S. S. Maram et al.

(in this context: the usage of literal characters from established belief or mythology) create a pragmatical upper cap on the number of characters a designer can draw from, but the adulteration of actual religious figures can arguably bear potential for controversy and cultural offense. There have been numerous controversies over transplanting gods into video games, such as Hanuman: The Boy Warrior [4], where the literal depiction of an important Hindu figure was perceived as nothing but denigration; or Faith Fighter [26], where multiple gods from world religions compete in brawl combat and especially Muhammad had to be censored, as Islamic belief prohibits the depiction of gods. These controversies revolve around how certain religious groups find the concept of “controlling gods through a joy-stick” or them “being inflicted damage” as offensive [1,2,43]. It becomes essential to realize that creating immersive experiences can not come at the cost of cultural sensitivity. This urges the need for a set of regulations or a framework for designers that allows drawing inspiration for characters from mythology without transplanting the worshipped iconography and twisted narrative. The contributions of this paper include: – A three-step framework that allows game designers to prevent transplanting gods from mythology and support the creation of new yet culturally sensitive characters. – An implementation of the proposed framework on Hindu mythology that proves the framework’s efficacy as well as the publication of a taxonomy of game mechanics inspired from Hindu mythology.

2 2.1

Related Work Cultural Affordances in Games

Social scientists view games as a new form of cultural expression, and collective behavior [10,18]. Seif El-Nasr et al. specifically expound how even video games are subjected to cultural and religious perception [15]. Convincing narratives and the diminishing line between graphics and reality have made games successful in influencing people [21]. This diminishing line makes it important to study cultural affordances in games. We pay attention to religion since it has been established how video games can create reflection on understanding religion [10]. Twisting actual narratives for game mechanics might influence perceptions and alter religious facts for players [10]. Consider the game Hanuman the Boy Warrior [4]. The game requires Hanuman (protagonist) to explore a forest and fight demons. The character re-spawns with the message “You have failed” every time the character falls into water. This is a twist to religious beliefs where Hanuman is considered immortal and has extraordinary aquatic abilities. Vasalou et al. reinforce how cultural sensitivity is essential in Serious games and mentions the importance of culturally authentic narratives [44]. The impetus of this work is to help designers design novel and culturally sensitive characters using elements from mythology as inspiration.

AstraVerse

2.2

5

Theology and Gaming

Ferdig mentions how religion can either be found explicitly or implicitly in games [17]. Above that, interactions between religion, new media and games have been discussed frequently [6,13,20,33,34,38]. Radde-Antweiler et al. and Detweiler provide a detailed account of how historically, various theologists have studied the influence of new media and games on how religion is being shaped in the modern world [13,33]. Vsisler’s work studies the perception of the Arab region, and how Islamic communities are perceived by players due to their representation in video games [38]. Their findings show how representation in games, unfortunately, has stereotyped communities to players and emphasizes how cultural and theological representation is an important point of discussion. Most work in religion and games has revolved around studying the effects of video games on religious perception. To our knowledge, there is no discussion on how religion can inspire designers develop new mechanics and characters, which is a gap this paper addresses. 2.3

Game Design Frameworks and Taxonomies

Taxonomies provide a way to organize and classify themes and concepts [3]. Researchers have established taxonomies on various concepts such as game mechanics [29], death and rebirth [11], bugs [22], player modelling [39], games for health [14,24], platformer games [40]. Frameworks for Serious Game design [8], Motion-Based Games [27], Collaborative Games [23] discuss design guidelines for different classifications of games. Previous work constituted discussions on formal design procedures and rules but does not discuss religion and cultural affordances. This work offers a case study on how the proposed taxonomy helps identify game mechanics from a specific culture and uses these mechanics to help designers build interesting yet appropriate characters and narratives.

3

AstraVerse Framework

This framework can be broken down into three steps as follows: – Mythological Derivation: Establishing the taxonomy of mythological references. – Generative Step: Facilitating the creation of new characters using the taxonomy. – Evaluative Step: Evaluating generated characters on (a) creative and (b) cultural scales. 3.1

Mythological Derivation

The goal of the first step is to compile a taxonomy based on the particular mythological background. In order to develop this taxonomy, we first identify various game elements that might benefit from this mythological influence [37].

6

S. S. Maram et al.

These primarily boil down – but are not limited – to character visualization, narrative, character abilities, in-game items or collectables, exploration, and combat mechanics. From the chosen mythology, we refer to cultural literature, internet threads and individual experts in mythology to identify popular mythological elements of interest. Finally, we construct nomenclature and visual iconography that can inspire game mechanics as well as characters and other elements. 3.2

Generative and Evaluative Steps

The generative step aim at verifying if the developed taxonomy allows the creation of characters. Designers – independent from the constructors of the taxonomy – are presented with the taxonomy and are requested to create new characters (participatory design). The evaluative step then verifies if the newly developed characters are deemed as creative and culturally valid. A feasible audience to review these with regards to creativity and novelty might arguably be senior game designers. The cultural validity however should be verified by performing qualitative and quantitative studies using participants that identify with or are familiar with the cultural backdrop associated with the particular mythology.

4

Applying the Framework on Hindu Mythology

The backlash certain games received from Hindu religious groups [1,2,43] renders Hindu mythology an appropriate benchmark to verify if characters inspired from Hindu mythology can create engaging characters while still being culturally valid. In order to showcase and evaluate a use case taxonomy (c.f. Sect. 4.1) and investigate the general capabilities of the proposed framework (c.f. Section 3), we organized a participatory design workshop. The participants of this workshop were introduced to the taxonomical elements outlined in Tables 1, 2, 3 and 4 with visual guiding of characters from Figs. 1 and 2. After that, they were requested to design a narrative and describe as well as sketch the visualized game character(s). The following section tackles taxonomical elements constructed from the framework (step 1 of the framework), whereas the subsequent section presents designers creating new characters using the taxonomy (step 2 of the framework) and finally qualitative remarks and interpretations from senior game designers (step 3(a) of the framework) about these creations, as well as assessments of cultural and religious appropriateness from a population with Hindu background (step 3 (b) of the framework). 4.1

Mythological Derivation [Step 1]: Establishing a Taxonomy on Mythological Hindu Characters and Mythical Objects

Hindu mythology discusses and presents a wide range of gods. We refer to works from Bansal [5] and Hindu epics such as the Mahabharata and Ramayana to identify characters and artifacts. Apart from these, we utilized popular illustrated

AstraVerse

7

mythological comics such as Amar Chitra Katha, internet forums as TvTropes [28], and popular Hindu Mythology YouTube channels such as KidsOne to aid building the taxonomy. The attributes, narrative and game mechanics these avatars inspire have been separated from their nomenclature and visual iconography, presented in Tables 1, 2, 3 and split into groups according to the classification by Parrinder [32]. We also identify elements which help the narrative flow and complete the core game loops. These may entail ways to heal characters, collectables or ways to transport around locations/maps involved in the narrative. Eventually, we derive a collection of mythological objects from Hindu mythology and how they could be used in game loops (c.f. Table 4). 4.2

Generative Step [Step 2]: Facilitating the Creation of Characters from the Established Hindu Taxonomy

For the participatory design workshop, we recruited three amateur game designers from a game design program (Masters degree) at . None of the participants identified as Hindus, and only one of them claimed to be partially aware of Hindu mythology. The following procedure was used in the workshop: – Take informed consent and introduce them to the context of the workshop i.e. to generate game narratives, design, sketch, or narrate visualized game characters using a taxonomy. – The researcher educates the participant on the developed taxonomy in Sect. 4.1. – With the introduced knowledge and ideas, the participant is requested to design and visualize game characters. 4.3

Constructed Outcomes from Generative Step

This section comprises the outcomes (transcribed from narration) of the Participatory Design from participants P1, P2 and P3 in terms of narrative creations and possible visual representations of these, derived as concept art during the workshop. P1: “A Cosmic Time Portal”. – Narrative: “A hiker slips down the mountain and falls into a celestial portal connecting the ancient times. He falls down to the armoury of celestial weapons. He is thrown back to the current day with the weapons stuck to him forever. The portal also throws ancient Asuars and powerful demons back into the modern day setting” – Visualization : “In my game I would like to build a massive female villain. I might have a three headed villain like Ravan (Row 17 Table 2) and have a extreme size like Kumbakaran (Row 18 Table 2). I will also arm her with Nature Powers like Krishna (Row 8 Table 1). Each head controlling different aspects of nature”. A visual representation is presented Fig. 3.

8

S. S. Maram et al. Table 1. Understating attributes and characteristics of dashavatara

Sr no. Avatar name Avatar description

Genre of narrative

Usable game mechanics

The Serial Number in the table indicates the Visual representation in Fig. 1 1

Matysa

In the zoomorphic version the avatar is a fish with the horn. In the anthropomorphic version the upper half is human and the lower half resembles a fish

Saves creation from a great flood. (Similar to flood myths across cultures), Recovers ancient scriptures by defeating a horse headed demon

Underwater speed due to fish fins, Underwater strength, Mythical horn

2

Kurma

In the zoomorphic version the avatar is a turtle. In the anthropomorphic version the upper half is human and the lower half resembles the shell of a turtle

Carries the weight of the world on the shell, Acts as a axis in mythology to churn a mythical cosmic ocean

Underwater agility due to turtle fins, Extreme focus and stability, Extreme strength, Hard protective shell

3

Varaha

The Varaha avatar has a boar head and a human torso as lower half

Goes to the depth of a cosmic ocean to lift Goddess Earth with tusks. Defeats fierce demons who imprison Goddess Earth

Agility in both water and land, Extreme speed, Mythical and sharp tusks, Boar face, Four hands

4

Narasimha

The avatar is visualized with a Lion head and human lower half

The avatar defeats a smart demon with mythical abilities

Extreme strength, Lion face, Lion like sharp claws and sharp teeth., Four hands, Carries a mace, Cosmic conch and Rotating sharp disk

5

Vamana

This avatar is visualized as a dwarf carrying The avatar stops a demon king from an umbrella and jug of water performing supreme sacrifices and restore the heavens to the Hindu King of Gods

6

Parushurama This avatar is visualized as a powerful Saint The avatar is credited to bring order and peace by defeating unjust kings who were ruling land with tyranny and greed

Fierce warrior, Yields the axe, Skilled in arms, Knowledge on how wield to celestial weapons, Extreme aggression, Skilled in rituals and academic knowledge

7

Rama

This Avatar is visualized as a Prince

The avatar is credited of defeating a powerful 10 headed demon Ravana. This avatar is the central protagonist of the Hindu Epic “Ramayana”

Skilled archer, Discipline, Kind, Academically strong, Celestial bow, Celestial and divine arrows

8

Krishna

This avatar is visualized as a Prince

The avatar defeats his Uncle, The avatar is a pivotal character of the Hindu Epic “Mahabharata”

Skilled warrior, Yields the rotating cosmic flame disc, Yields celestial weapons and shields, Academically strong, Mutable in size, Mythical powers, Great rapport with animals and nature, Multiple arms, Master flute player

9

Buddha

This avatar is visualized a monk

The primary purpose of this avatar was to teach morals and principles

Academically strong, Great orator, Peace loving, Calm

10

Kalki

A person riding a mythical white horse and carrying a long sword

The avatar is supposed to eliminate human human race and restart cycle of life

Swift, Skilled swordsmen, Mythical horse, Flaming sword

Mutable in size(Dwarf to Giant), Mythical powers (Chants, Spells), Carries a protective umbrella, Magic water

Table 2. Understanding Attributes and Characteristics of Characters from the Mahabharata and Ramayana Sr no. Avatar name Avatar description

Genre of narrative

Usable game mechanics

The Serial Number in the table indicates the Visual representation in Fig. 1 11

Arjuna

This avatar is visualized as a prince

The son of the King of gods. He is the best archer in Hindu Mythology. He yields all celestial weapons. There are instances where Arjuna is also portrayed as a eunuch

Plethora of celestial weapons, Infinite arrow quiver, Cosmic bow, Indestructible chariot

12

Karna

This avatar is visualized as a prince

The son of the Sun god. Karna is a great archer second to Arjuna. Karna is considered to be radiant and bright as the sun

Indestructible golden armor, Radiant earrings reflecting Solar radiation, Cosmic bow, Loyalty, Recipient of many curses

13

Bheema

This avatar is visualized as a prince

The son of Wind God. Has the strength of 1000 elephants. Feared among Giants and Asuras

Extreme strength, Yields of the cosmic mace, Pride in strength, Affection towards loved ones

14

Duryodhana This avatar is visualized as a prince

15

Gatotkach

This avatar is visualized as a friendly demon The son of Bheema (Row 13). Is a mighty warrior with magical powers

16

Hanuman

Has anthropomorphic iconography. Avatar is visualized with a Monkey head, human lower half and has a Monkey Tail

The avatar is the son of the wind god. Student Fierce warrior, Yields the mace, Extreme of the Sun god and the principle associate of strength, Extreme loyalty, Capable of flight, Rama (Row 7 Table 1). Blessed as one of the Skilled in academic knowledge immortals in Hindu mythology

17

Ravana

This Avatar is visualized as the king of Sri Lanka with ten heads

He is the central antagonist of the Hindu Epic “Ramayana”. Has his soul in his navel, making him invincible at the rest of his body. Mighty with 10 heads and has knowledge to many celestial weapons

Skilled warrior and fierce commander, Extreme pride and lust, Great devotee of certain Gods, Access to celestial and cosmic weapons, Academically strong, Posses aerial flight machines

18

Kumbakaran This avatar is visualized as an extremely large giant

He is a massive asura who is known to squash enemies to pulp. He is slow and has ferocious appetite. Sleeps for 6 months of the year and awake the other 6 months

Extreme strength, Slow maneuver, Can be only hit by celestial weapons, High Energy drain

19

Shiva

Is labelled as the god of destruction. His dark Third eye whose opening leads to annihilation, blue color is attributed to carrying a dangerous Trident, Hour glass shaped drum which causes poison in his throat to save humanity cosmic vibration, Serpent neck, Commander of God Soldiers, Lunar Control

Is one of the supreme Gods in Hindu Mythology. Has a serpent around the neck and the moon in the form of a crescent locked in the hair

Prime antagonist in the Mahabharata. Is know Extreme pride and cunning, Iron body in the for his extreme greed. Is a mighty warrior and upper half, Fragile Lower half the body, ranks among the greatest wrestlers Extremely experienced wrestler, Death grip Magical illusions, Mutable size, Multiplied powers at nightfall

AstraVerse

9

Table 3. Understanding attributes and characteristics of gods of various qualities and elements Sr no. Avatar name Avatar description

Genre of narrative

Usable game mechanics

The Serial Number in the table indicates the Visual representation in Fig. 2 20

Saraswati

This avatar is visualized as a goddess

The goddess is treated as the goddess of education

Music powers (Carries the instrument Veena), Purifying powers, Highest educational knowledge, Resides in a white lotus, Travels in a white swan

21

Lakshmi

This avatar is visualized as a goddess

The goddess is the spouse of Lord Vishnu (source of Table 1) and is has incarnations along with the Dashavatara (Table 1)

Giver of wealth, fortune, Power of maya (“illusion”), Giver of agriculture, fertility, health, courage

22

Bhoodevi

This avatar is visualized as a woman

This goddess is treated as Mother Earth

Strength (carries the weight of living beings), Earth control powers (rotation, tunnels), Landmass control (earthquakes, avalanches)

23

Durga

There have been many popular representation of this goddess These range from two arms, to 10 arms carrying various weapons

This goddess is considered one of the most powerful deity in Hindu mythology

Has a lion as a vehicle, Carries discs, mace, bows, swords, conch in certain representations, Creates powerful Illusions

24

Surya

This avatar is a radiant human

This god is treated as the Sun God

Solar flames, Extreme gravity, Controls day and night, Extreme speed (Rides a chariot with 7 horses)

25

Vayu

This avatar is visualized the as a human with Wind capabilities

This god is the controller of Air, Wind and gases

Flying abilities and Extreme speed, Can control winds (Storms, Tornado’s)

26

Agni

This avatar is visualized as a human engulfed in flames

The God Agni is the representation of Fire

Symbolically represents Fire,Receives damage from rain and aqua avatars, Huge appetite (burns down forests), Guardians of Divine weapons

27

Varuna

This avatar is visualized the as a human with Aqua capabilities

This god is the king of the oceans and aquatic Extreme underwater abilities Commander of the life Sea animals Ocean Control (Tsunamis, Cyclones, Gateways)

28

Ganesha

God with a human torso and Elephant head This is the god on removing all obstacles and education

Strength, Devotion and high patience, Magical elephant tusks and elephant trunk, Slow Movement, A jewellery adorned Rat as a transport vehicle

29

Garuda

He has a human body and face. The mouth Garuda has been discussed in Hindu is modified to the shape of an eagle. Garuda mythology on multiple occasions. Garuda is has a huge wingspan the vehicle of Lord Vishnu for most of his Avatars (Table 1)

Aerial ability, Cosmic speeds, Large wingspan, Commands over eagles, Consumes snake as prey, High loyalty

Table 4. Understanding Mythical Objects and their Role in Game Loops Sr no. Game loop element

Item name

Abilities

1

Shields

Karan Kavach, Shiv Kavach, Bramha Kavach

The Karan Kavach is the Armour given to Karna (Row 12 Table 2) by the Sun god. The armour is a supreme armour and the one who wears it is invincible at battle. The Shiv Kavach is the armour of the gods. The Bramha Kavach could only be destroyed by the BramhaAstra (Row 6 Table 4). Shields can be introduced in game loops as collectables whose effects span over a time period. Shields can also be introduced as inbuilt character abilities with a cooling period once used

2

Flags

Garuda Flag, Indra Flag, Hanuman Flag, Kaama Flag

Flags act as collectables in a game loop. These collectables can act as short term abilities. The Garuda Flag is a reflection of speed, The Indra flag is a symbol of authority, The Hanuman Flag is powerful flag which resits incoming celestial weapons, Tha Kaama Flag is the flag of love

3

Healing Herbs and Elixirs Sanjeevani, Amrit

It is common for protagonists and other character during game play to loose health. Healing herbs such as the Sanjeevni and Amrit have narrative references in Hindu mythology to restore life and health

4

Transport

Garuda, Surya Vimana, Hansa Vimana, Pushpaka Vimana, Tripurajit Vimana

The term Vimana is analogus to flight. Hindu mythology had references to Vehicles which broke the time space barrier. The Surya vimana is a golden Chariot with 7 White horses (each corresponding to one day of the week). The Hansa Vimana is a White chariot guided by swans. The Pushpaka Vimana is a chariot of Ravana (Row 17 Table 2). The Tripurajit Vimana is cosmic Chariot used by Shiva (Row 19 Table 2). Garuda is a Eagle with a huge wingspan capable of flying through the cosmos

5

Elemental Weapons

BhumaAstra (Earth), AgniAstra (Fire), VayuAstra (Wind), VarunaAstra (Water)

The BhumaAstra is a weapon capable of shattering the earth and digging tunnels. The AgniAstra is a fire emitting weapon. The VayuAstra is told to travel at great speeds. The VarunaAstra is capable of unleashing large water bodies on armies, fires

6

Cosmic Weapons 1

Bramhastra, Trishul, Sudarshana Chakra, Vajra

The Bramhastra is an extremely powerful weapon capable of destroying armies and cities. The Trishul is the weapon of Shiva (Row 19 Table 2). Visually similar to the trident. Sudarshana Chakra is a disc which travels at cosmic speeds to behead enemies and return to the owner once done. The Vajra is analogus to the mighty thunderbolt

7

Cosminc Weapons 2

PashpataAstra, NarayanaAstra, BrahmashirshaAstra, RudraAstra

The PashupataAstra is one of the irresistible weapons, which requires high skill to use. The NarayanaAstra is a weapon which showers weapons from the sky upon enemies. The shower can only be stopped once the enemy bows to the power of the weapon and disarms themselves. The BrahmashirshaAstra is an advanced version of the BramhaAstra and is said to cause four times more damage. The RudraAstra translates to the Furious Weapons, it is the only weapon which can counter a BramhaAstra

8

Cosmic Weapons 3

NagaPasha, GarudaAstra, Sammohanastra, Prajnastra

The NagaPasha is a weapon dedicated to the Snake Gods. The weapon releases serpents to attack the enemies. This can be countered by the GarudaAstra, which summons Garuda (Row 4 Table 4) the eagle, which sweeps the serpents. The SammohanaAstra intoxicates enemies, while the PrajnaAstra recovers them from the intoxication

8

Cosmic Bows

Gandeeva, Sharanga, Vijaya, Pinaka

Not all bows can handle Cosmic Weapons. These are a few bows have been used by various characters in Hindu mythology. The typical Characteristic of these bows include high string tension, Multiple strings, made from celestial wood from alternate dimensions

9

Cosmic Swords

Nandaka, Chandrahas

The Nandaka is the sword of the Dahsavatars. The Chandrahas translates to “crescent” in Hindu mythology there has been references where the crescent of the Moon acts as a sword for the Gods

10

S. S. Maram et al.

Fig. 1. Visual representation of characters discussed in Table 1 and Table 2.

Fig. 2. Visual representation of characters discussed in Table 3

Fig. 3. Character visualiza- Fig. 4. Character visualiza- Fig. 5. Character visualization of participant 1 tion of participant 2. tion of participant 3.

P2: “Mystique Musician”. – Narrative: “The narrative is linked to African tribes and their freedom struggle against colonial empires. A young girl has music has her genre of attacks. She is a black musician who evolves as levels in the game progress. She yields all mythical instruments mentioned in the taxonomy (Row 20 Table 4, Row 8 Table 1 and Row 19 Table 2).” – Visualization: “I am thinking of giving the protagonist multiple arms like Narasimha (Row 4 Table 1) and Varaha (Row 3 Table 1). The character will unlock more arms and more weapons as the levels progress”. A visualization of this character is shown in Fig. 4 P3: “War of the Elements”. – Narrative. “It is important that people are bought aware of climate changes. In my game I want to make use of Natural elements in Hindu Mythology such a Lunar control, Sun energy and Wind energy to show players how the lack of either of them leads to trouble. I want characters in my game to control nature”

AstraVerse

11

– Visualization. Like Shiva (Row 19 Table 2) has a moon in his forehead. The crescent of the moon can act as a sword facing terrestrial enemies. While fighting aqua characters the moon would glow to be a dull disc to cause low tide. A negotiating character would be a character with earth capabilities (Row 22 Table 3) or solar capabilities since they would have a greater gravitational pull.” A Visual Representation is presented in Fig. 5. 4.4

Creative Evaluation [Step 3(a)]: Evaluating the Creative Dimension of Workshop Outcomes

Procedure. First, the produced narratives and characters (c.f. Section 4.3, Sects. 4.3 and 4.3) were presented to two senior game designers to evaluate the creative and novelty element of the creations. We recorded these interviews and asked them to quantify their estimation as well. Measures. For measuring creativity and novelty, we asked the senior game developers to judge items (on a scale of 5) based on Brookhart’s rubric [7]. On top of this, the interviewees were asked to comment on the particular narratives qualitatively. Participants. In total, (na = 2) senior game developers (those with 3+ years of Game Production experience) assessed the creative quality of the creations. Results of Creative Evaluation. Table 5 contains the quantitative assessment of senior game designers towards the produced characters. Furthermore, we asked them to elaborate on their scores. Judge 1, who scored Participant 2 the highest (in terms of cumulative points compared to other participants), shared “The combination of mythical influence on historical freedom wars is fascinating, ranging from colonial invasions, musical warfare and finally introducing mythical physical characteristics. The other idea (Participant 1) had mythical power from both ends, while here it is more the mythical powers of the hero taking on huge armies in different conditions. ” Judge 2 had appreciative feedback on Participant 3’s work - “I really like the idea of a mutable moon sword causing tides and that being the primary way it fights. That feels really cool. The character also being made of bushes and trunk as hair is also an interesting combination. ” 4.5

Cultural Evaluation [Step 3(b)]: Evaluating the Cultural Validity of Workshop Outcomes

Procedure. Students with a Hindu background were presented Fig. 3, 4, 5 as well as depictions1 of a Hindu goddess that raised controversial discussions in 1

A reference to the image depicting the representation of Kali from the game SMITE can be found at https://images.news18.com/ibnlive/uploads/2012/07/kalismite.jpg.

12

S. S. Maram et al.

the multiplayer online battle arena game SMITE [42]. Participants were asked to comment on the representation in SMITE as well as on the images produced by our taxonomy qualitatively, which was later classified by structuring content analysis. Measures. Regarding the cultural assessment, participants were surveyed using semi-structured interviews on their appraisal and statements towards both visualizations. Participants. In total, (nb = 39) students were recruited through a call for participation from online Indian Communities at for the cultural appropriateness evaluation. From the recruited student participants, 92.9% identified as Hindus (3.6% Atheists and 3.6% Jains). All of the participants claimed to be aware of Hindu mythology and the most of them also knew other mythologies, most popularly Greek (69%). Table 5. Quantitative Evaluation [7] of Designed Characters and Narratives in Sect. 4.3 by Experienced Game Designers Judge 1

Judge 2

P1 P2 P3 P1 P2 P3 Variety of ideas and contexts

3

5

3

3

3

4

Variety of sources

4

4

3

4

3

3

Combining ideas

4

5

4

4

3

4

Communicating something new 3

5

4

3

3

3

Results of Cultural Evaluation. Participants were shown a representation of Goddess Kali from the game SMITE. The Hindu-affiliated participants had mixed opinions on the representation with 10% of them being “offended ” by the portrayal. Another 10% of the participants claimed the representation is “Perfectly normal ”. Around 20% of them were not offended but claimed “they were not comfortable” with such a visualization. The majority of 53% claimed “they might be okay with such a representation, but know people who would be offended ”. Also, participants qualitatively reinforced that (P9): “Kali presented like this, people here in might find it amusing. If you are planning to launch in India, be ready for controversies and only criticism. ” After presenting SMITE’s representation of Goodess Kali, we asked participants to comment on characters from Fig. 3, 4, 5. For Fig. 3, 76% of the students felt the characters were “Perfectly Normal” and 10% of the students shared “they might be okay with such a representation, but people who would be offended ”. A 5% reported “they are not comfortable” and 3% of the participants felt the representation was offensive. With reference to this, (P19) interpreted the result that

AstraVerse

13

Fig. 6. Quantitative results reflecting cultural acceptance of characters from SMITE and generated characters as part of the study shown in Fig. 3, 4, 5

“Three heads makes me think of Ravana but he had ten [heads]. Also Reminds me of Lord Brahma with his multiple heads but it doesn’t strike as an instant connection.” For Fig. 4, 82% of the participants felt the characters were “Perfectly Normal ” and only 5% shared “they might be okay with such a representation, but know people who would be offended ”. No participant responded to be “offended ” and only 7% of the participants shared “they were not comfortable”. In reference to Fig. 4, (P2) declared: “I find this art more artistic rather than something offensive. Even if it’s explicitly implying that the art has taken some elements from Hindu mythology, I still find it creative and something joyful, a person who is enjoying music. I can’t see how this art is gonna be offensive to people.” For Fig. 5, 79% of the participants felt the characters were “Perfectly Normal ” and 12% of them shared “they might be okay with such a representation, but people who would be offended ”. 5% reported “they were not comfortable”. No participant responded to be offended. (P6) added that “healing herbs reminds me of Sanjeevani from Ramayana. There are many mythological aspects here, like the hair represents vegetation and the solar shield representing the Surya Vamsha.”

5

Results and Discussion

Fig. 6 reveals an arguably positive trend in cultural acceptance of characters when built using the provided framework. To investigate this effect, we calculated statistical significance utilizing Wilcoxon rank sum tests on the ordinally ranked response categories between the SMITE example and each of the characters produced within our workshop [47]. For measuring the impact of these

14

S. S. Maram et al.

comparisons, we additionally computed effect sizes r after Rosenthal [35]. As hypothesized, the SMITE portrayal of Kali produced significantly higher indications of offensiveness than the characters of Fig. 4 (p < 0.05, r = 0.79), Fig. 5 (p < 0.05, r = 0.83) and Fig. 3 (p < 0.05, r = 0.74); all showing large effect sizes. Even if the perceived creativity and novelty of characters are always subjective and dependent on the target population that would play such a game, senior game designers approved the potential of the taxonomy to construct innovative characters by means of inspiration and combination. SMITE’s version of Kali raised considerably more rejections than endorsement. This highly contrasts the feedback that characters produced by our taxonomy received, with most of the responses judging all three of them as perfectly normal, none or only single mentions of offense and only minor doubts of the appropriateness for peers within the Hindu community. Statistical significances with large effect sizes throughout all comparisons further strengthen the potential of our approach.

6

Future Work and Conclusion

In the next iteration of the work, we plan to use standard UX research methods such as Grounded Theory and Card sorting to create the taxonomy. We plan to have a larger participatory design workshop for creating characters (to eliminate any creative bias). In the future, we seek to come up with metrics to determine upper limits on how much a designer can borrow from a taxonomy and prevent misuse by accidentally creating characters which might be perceived as culturally insensitive characters. In this paper, we presented a framework which allows designers to draw inspiration from mythological references while maintaining cultural validity. As a case study, we build a taxonomy using the proposed framework over Hindu mythology to help designers create culturally sensitive characters. To showcase that the framework supports the creation of creative and novel characters, we discussed outcomes from a study where game designers used a taxonomy constructed from the framework to build novel characters. These created characters were evaluated on the lines of creativity and cultural sensitivity. The outcomes of these evaluations supported our hypothesis of allowing designers to create creative, culturally inclusive and immersive characters using the proposed taxonomy. We believe, the scope of the is paper is not limited to Hindu mythology and is equally valid for other mythologies. Acknowledgments. The research team would like acknowledge and thank all the participants and senior designers for taking part in our study. We would also like to acknowledge and thank the University of California, Santa Cruz for their support.

References 1. Religious groups protest game for its depiction of kali- technology news, firstpost (2012). https://www.firstpost.com/tech/news-analysis/religious-groupsprotest-game-for-its-depiction-of-kali-3603945.html

AstraVerse

15

2. Alexander, L.: Hindu statesman criticizes sony’s hanuman: Boy warrior (2009). https://www.gamasutra.com/php-bin/news index.php?story=23280 3. Antle, A.N., Wise, A.: Getting down to details: using learning theory to inform tangibles research and design for children. Interact. Comput. 25(1), 1–20 (2013) 4. Aurona Technologies: Hanuman the Boy Warrior (2009) 5. Bansal, S.P.: Hindu gods and goddesses. Smriti Books (2005) 6. Bornet, P., Burger, M.: Religions in play: games, rituals, and virtual worlds, vol. 2. Theologischer Verlag Z¨ urich (2012) 7. Brookhart, S.M.: How to create and use rubrics for formative assessment and grading. Ascd (2013) 8. Buchanan, L., Wolanczyk, F., Zinghini, F.: Blending bloom’s taxonomy and serious game design. In: Proceedings of the International Conference on Security and Management (SAM), p. 1. The Steering Committee of The World Congress in Computer Science, Computer (2011) 9. Bunian, S., Li, K., Jemmali, C., Harteveld, C., Fu, Y., Seif El-Nasr, M.S.: Vins: Visual search for mobile user interface design. In: Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems, pp. 1–14 (2021) 10. Campbell, H.A., Wagner, R., Luft, S., Gregory, R., Grieve, G.P., Zeiler, X.: Gaming religionworlds: why religious studies should pay attention to religion in gaming. J. Am. Acad. Relig. 84(3), 641–664 (2016) 11. Cuerdo, M.A.M., Melcer, E.F.: “ i’ll be back”: A taxonomy of death and rebirth in platformer video games. In: Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems, pp. 1–13 (2020) 12. De Wildt, L., Aupers, S.: Playing the other: role-playing religion in videogames. Europ. J. Cult. Studies 22(5–6), 867–884 (2019) 13. Detweiler, C.: Halos and avatars: Playing video games with god. Westminster John Knox Press (2010) 14. Dormann, C.: Toward ludic gerontechnology: a review of games for dementia care. DiGRA/FDG (2016) 15. El Nasr, M.S., Al-Saati, M., Niedenthal, S., Milam, D.: Assassin’s creed: A multicultural read. Loading... 2(3) (2008) 16. Evangelopoulou, O., Xinogalos, S.: Myth troubles: an open-source educational game in scratch for Greek mythology. Simul. Gaming 49(1), 71–91 (2018) 17. Ferdig, R.E.: Developing a framework for understanding the relationship between religion and videogames. Online-Heidelberg J. Relig. Internet 5 (2014) 18. Grieve, G.P., Campbell, H.A.: Studying religion in digital gaming. a critical review of an emerging field. Online-Heidelberg J. Relig. Internet 5 (2014) 19. Hagen, U.: Where do game design ideas come from? invention and recycling in games developed in sweden. In: DiGRA Conference (2009) 20. Heidbrink, S., Knoll, T.: Online-Heidelberg journal of religions on the internet/5 religion in digital games: multiperspective and interdisciplinary approaches. Online-Heidelberg J. Relig. Internet (2014) 21. Von der Heiden, J.M., Braun, B., M¨ uller, K.W., Egloff, B.: The association between video gaming and psychological functioning. Front. Psychol. 10 1731 (2019) 22. Lewis, C., Whitehead, J., Wardrip-Fruin, N.: What went wrong: a taxonomy of video game bugs. In: Proceedings of the fifth international conference on the foundations of digital games, pp. 108–115 (2010) 23. Loparev, A., Egert, C.A.: Toward an effective approach to collaboration education: A taxonomy for game design. In: 2015 IEEE Games Entertainment Media Conference (GEM), pp. 1–4. IEEE (2015)

16

S. S. Maram et al.

24. McCallum, S., Boletsis, C.: Dementia games: a literature review of dementia-related serious games. In: Ma, M., Oliveira, M.F., Petersen, S., Hauge, J.B. (eds.) SGDA 2013. LNCS, vol. 8101, pp. 15–27. Springer, Heidelberg (2013). https://doi.org/10. 1007/978-3-642-40790-1 2 25. Mete, F.: The creative role of sources of inspiration in clothing design. Int. J. Cloth. Sci. Technol. 18(4) 278–293 (2006) 26. Molleindustria: Faith Fighter (2009) 27. Mueller, F., Isbister, K.: Movement-based game guidelines. In: Proceedings of the Sigchi Conference on Human Factors in Computing Systems, pp. 2191–2200 (2014) 28. Multiple: Hindu mythology / myth, https://tvtropes.org/pmwiki/pmwiki.php/ Myth/HinduMythology 29. Nacke, L.E., Grimshaw, M.: Player-game interaction through affective sound. In: Game sound technology and player interaction: Concepts and developments, pp. 264–285. IGI global (2011) 30. Ninja Theory: Hellblade: Senua’s Sacrifice (2017) 31. Olesen, J.F.: Design processes in game jams: Studies of rapid design processes. In: Extended Abstracts Publication of the Annual Symposium on Computer-Human Interaction in Play, pp. 723–726. CHI PLAY ’17 Extended Abstracts, Association for Computing Machinery, New York, NY, USA (2017). https://doi.org/10.1145/ 3130859.3133226 32. Parrinder, E.G.: Avatar and incarnation (1982) 33. Radde-Antweiler, K.: Religion is becoming virtualised. introduction the the special issue on religion in virtual worlds. Online-Heidelberg Journal of Religions on the Internet: Vol. 03.1 Being Virtually Real? Virtual Worlds from a Cultural Studies’ Perspective (2008) 34. Radde-Antweiler, K., Waltmathe, M., Zeiler, X.: Video gaming, let’s plays, and religion: The relevance of researching gamevironments. Gamevironments (1) (2014) 35. Rosenthal, R., Cooper, H., Hedges, L., et al.: Parametric measures of effect size. Handbook Res. Synth. 621(2), 231–244 (1994) 36. Sbai, O., Elhoseiny, M., Bordes, A., LeCun, Y., Couprie, C.: DesIGN: design inspiration from generative networks. In: Leal-Taix´e, L., Roth, S. (eds.) ECCV 2018. LNCS, vol. 11131, pp. 37–44. Springer, Cham (2019). https://doi.org/10.1007/9783-030-11015-4 5 37. Sicart, M.: Defining game mechanics. Game Studies 8(2), 1–14 (2008) ˇ 38. Sisler, V.: Digital arabs: representation in video games. Europ. J. Cult. Studies 11(2), 203–220 (2008) 39. Smith, A.M., Lewis, C., Hullett, K., Smith, G., Sullivan, A.: An inclusive taxonomy of player modeling. University of California, Santa Cruz, Tech. Rep. UCSC-SOE11-13 (2011) 40. Smith, G., Cha, M., Whitehead, J.: A framework for analysis of 2d platformer levels. In: Proceedings of the 2008 ACM SIGGRAPH Symposium on Video Games, pp. 75–80 (2008) 41. Supermassive Games: Until Dawn (2015) 42. Titan Forge Games: Super Metroid (2019) 43. Usher, W.: Smite offends hindus, catholics, jews, with porno-style depiction of kali (Jul 2012), https://www.cinemablend.com/games/SMITE-Offends-HindusCatholics-Jews-With-Porno-Style-Depiction-Kali-44645.html 44. Vasalou, A., Khaled, R., Gooch, D., Benton, L.: Problematizing cultural appropriation. In: Proceedings of the first ACM SIGCHI Annual Symposium on ComputerHuman Interaction in Play, pp. 267–276 (2014)

AstraVerse

17

45. Vigil Games: Darksiders (2010) 46. Wallace, S., et al.: Sketchy: drawing inspiration from the crowd. Proc. ACM Human-Comput. Interact. 4(CSCW2), 1–27 (2020) 47. Wilcoxon, F.: Individual comparisons by ranking methods. In: Kotz, S., Johnson, N.L. (eds) Breakthroughs in Statistics Breakthroughs in statistics, pp. 196–202. Springer (1992). https://doi.org/10.1007/978-1-4612-4380-9 16 48. de Wildt, L., Apperley, T.H., Clemens, J., Fordyce, R., Mukherjee, S.: (re-) orienting the video game avatar. Games Cult. 15(8), 962–981 (2020)

Comfortably Numb? Violent Video Games and Their Effects on Aggression, Mood, and Pain-Related Responses Gary L. Wagener(B)

and André Melzer

University of Luxembourg, 11 Prte des Sciences, 4366 Esch-sur-Alzette, Luxembourg [email protected]

Abstract. In contrast to findings that violent video game (VVG) exposure has a desensitizing effect on empathy and physiological reactivity to scenes of violence [1], no desensitization was found for player responses to pain stimuli in three lab experiments. Compared to a non-violent game, VVG exposure neither affected physiological responses, nor participants’ self-reports of perceived pain caused by thermal stress. In addition, the level of game immersion did not affect pain perception, pain tolerance, or aggressive behavior (study 3). In contrast, violent game preference was associated with lower reports of perceived proximal pain, distal pain, and greater antisocial behavior. However, all studies confirmed the detrimental effect of VVG on emotion: participants reported lower positive and greater negative affect after playing the violent compared to the nonviolent game. In sum, the present findings speak against a generalized desensitization effect of VVG on the player. Rather, our findings further support the notion of pain and pain-related responses as complex and multidimensional, modulated by individual, physiological, and contextual factors [2]. Keywords: Violent video games · Desensitization · Pain · Pain perception · Mood

1 Introduction The impressive economic and social success of video games as well as their potential effects has put the medium under extensive scientific scrutiny. In fact, whether violent video games (VVG) cause negative effects on player emotion, cognition and behavior has led to fierce debates among scholars [3–8]. In the light of the popularity and availability of video games (VG), the great interest of parents, stakeholders, and the general public about their potential effects, but also regarding current discussions about stable and consistent findings in psychology, replications, and meaningful advancements of existing approaches are inevitable to overcome simplifications and over-generalization of results on the effects of VVG.

© IFIP International Federation for Information Processing 2022 Published by Springer Nature Switzerland AG 2022 B. Göbl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 18–38, 2022. https://doi.org/10.1007/978-3-031-20212-4_2

Comfortably Numb? Violent Video Games and Their Effects

19

1.1 Violent Video Games and Aggression Desensitization According to the General Aggression Model (GAM; [9]), VVG exposure can have shortterm and long-term effects on aggression. Short-term exposure to VVG can influence a person’s affect, arousal, or cognition, increasing the likelihood that the person will behave aggressively [9, 10]. In contrast, long-term exposure to VVG may in turn influence individual factors, leading to a desensitization effect, increased trait aggression or the reinforcement of attitudes and beliefs towards aggression [10]. The present study addresses findings of a desensitizing effect of VVG exposure. Desensitization is characterized as a decrease in the psychological and physiological response to a stimulus after continuous exposure [11]. Compared to non-violent games, playing VVG has been reported to reduce physiological and/or neurological arousal as well as emotional responding [11–13]. However, desensitization caused by VVG is believed to affect internal processes as well as social interactions. Violence in video games has been reported to increase aggressive behavior, reduce feelings of empathy in the players, and increase aggressive affect [1, 10]. Regarding the latter, feelings like hostility, anger, and a sense for vengeance are increased after VVG exposure [10] which also negatively impacts mood, increasing negative affect and decreasing positive affect [14–16]. Regarding feelings of empathy, participants that were exposed to VG violence showed reduced automatic emotional reactions to harm befalling someone else [17, 18]. For children and adolescents, VVG exposure is assumed to increase the risk of desensitization, possibly affecting aggression levels and decreasing prosocial behavior [19]. 1.2 Violent Video Games and Pain Desensitization Another likely candidate for the desensitizing effects of violence in VG is pain. Pain denotes a complex concept that is modulated by individual, physiological, and contextual factors [2]. Moreover, pain has individual as well as social aspects. Based on the assumptions in the GAM, the potentially numbing and the emotion desensitizing effect of playing VVG should result in stimuli perceived as less painful, both for pain directed to the self (i.e., proximal pain) and above all pain observed in others (i.e., distal pain). The latter is also affected by reduced empathic responses after VVG exposure [1]. For example, studies have shown that exposure to VVG leads to increased desensitization, reducing physiological and emotional reactions to stimuli [10, 11]. The present study therefore tested whether playing VVG has a desensitizing effect on participants’ pain responses. Regarding the role of VG in improving health-related outcomes, a meta-analysis confirmed the pain distracting effect of playing VG [20]. To date, however, only few studies have tested the effects of playing VG on pain perceptions directly. In one study, the so-called cold-pressor test was conducted, in which participants hold their hand in ice-cold water for as long as possible while taking out paper clips [21]. The experience of immersion during gameplay was crucial for pain sensitivity (Study 2): Compared to solving a non-immersive puzzle game, having played a first-person 3-D game led to greater pain tolerance (as indicated by the greater number of paperclips retrieved from ice-cold water), as well as greater indifference towards people depicted as experiencing

20

G. L. Wagener and A. Melzer

displeasure. The authors attributed their findings to the desensitizing effect to pain in oneself and in others [21]. Stephens and Allsop [22] found that playing a VVG not only increased aggressive feelings and arousal (as indicated by heart rate), but also pain tolerance. Compared to a golf video game (i.e., the non-violent game), participants that had played a first-person shooter game found ice-cold water less painful (i.e., increased pain tolerance), as indicated by the longer time they held their hand in ice-cold water. The authors attribute this finding to the hypoalgesic effect of the emotional response that accompanies raised state aggression [22]. VVG exposure is thought to have a desensitizing effect not only on proximal pain but also on distal pain perception. In an event-related potential study [23], participants with no habitual experience in violent gaming showed reduced empathic responses to painful images after playing a VVG for 40 min. Participants with habitual violent gaming experience showed a desensitization effect to painful images already before gameplay. They also showed no additional decrease in empathic response to painful images after gameplay. The authors suggested that habitual violent gamers down-regulate their negative-emotional arousal to better perform in-game [23]. 1.3 Hypotheses Theoretically, then, players should become desensitized to aggression in their cognitive, physiological, and emotional responses after continuous exposure to VVG. For example, regarding desensitization in physiological response, Bartholow et al. reported that VVG exposure reduced event-related brain potentials which in turn predicted aggressive behavior in male adults, even when controlling for trait-aggression [11]. Therefore, three experimental studies tested the hypotheses that VVG exposure leads to reduced proximal (H1) and distal (H2) pain perception. In addition, VVG exposure is expected to increase pain tolerance (H3), increase aggressive behavior (H4), decrease physiological reactions to pain stimuli (H5), and decrease mood (H6). Study 1 tested these hypotheses in a between-subjects lab experiment. Study 2 used the same basic design, trying to replicate findings with different methods to increase validity. Study 3 introduced the additional factor of immersion into the study design.

2 Study 1 Study 11 tested the relationship between VVG exposure and proximal pain perception, aggressive behavior, and mood. It was hypothesized that in contrast to playing a nonviolent VG, playing a VVG decreases proximal pain perception, increases aggressive behavior, and decreases mood. 2.1 Methods Participants. Participants (N = 66; 50% females; M age = 22.92; SD = 2.88) were recruited at the University of Luxembourg. They indicated how much they played VG 1 All three studies presented were accepted for ethics approval at the University of Luxembourg.

Comfortably Numb? Violent Video Games and Their Effects

21

during typical weekdays, weekends, and holidays (1 = never; 2 = less than 1 h; 3 = 1–2 h; 4 = 2–3 h; 5 = more than 3 h). On average, participants played VG on a low to medium level (M = 1.83; SD = .99). Pain Perception, Pain Tolerance, and Aggressive Behavior. Pain was induced using the cold pressor test (CPT). A container (size in cm: 60 x 40 x 18) was filled with cold water (4 °C) controlled by the immersion cooler Julabo FT200. To measure pain tolerance, participants held their hand up to the forearm in the cold water for as long as possible. Time was kept using a stopwatch. To measure pain perception, participants verbally rated their pain level every 15 s using a numerical rating scale (0 = no pain to 10 = worst possible pain, [24]). To measure aggressive behavior, they assigned a minimum time requirement in the CPT for the next participant. For ethical reasons, the maximum duration for the CPT was set to three minutes. Target Games. Participants played both games on the Wii game console on a 46 flatscreen. For the violent condition, Manhunt 2 was chosen due to its high violence ratings. Participants controlled the game character, a patient who tries to escape from a closed psychiatric ward by killing people who get in his way. Participants in the nonviolent condition played Wii Sports Resort, which includes a variety of sports games. Both games were chosen as they were easy to play but challenging. Mood. Mood was assessed at two points in time with the positive and negative affect subscales of the PANAS [25] Each subscale comprises ten adjectives (e.g., active, guilty) that were rated on a 5-point scale (1 = not at all to 5 = extremely) on how strongly the person feels about each adjective. Internal consistency before gameplay was good for positive affect (Cronbach’s α = .82) and negative affect (α = .88). The second measurement was done immediately after gameplay. The items were given in a different, randomized order. Again, the internal consistency was good for positive affect (α = .83) and for negative affect (α = .89). Trait Aggression. To assess trait aggression, the Aggression Questionnaire [26] that includes the subscales for physical aggression (8 items), anger (6 items), and hostility (6 items) was used in its German version [27]. The items were rated on a 4-point scale (1 = I strongly disagree to 4 = I strongly agree). The combined scale for trait aggression showed good internal consistency (α = .80). Violent Video Games Preference. A novel five-item scale was designed for violent video game preference. Participants indicated how much they prefer and appreciate violent content in video games. Each item started with “I prefer games…” and were related to, among other factors, the motivation to intentionally behave antisocially in games (i.e., “…where I can hurt or kill others”), and to dominate others (i.e., “…that contain scenes of power and domination”). The scale had very good internal consistency (α = .85). Pain Sensitivity. To measure trait pain sensitivity, five items of the Pain Sensitivity Questionnaire (PSQ; [28]) were used. Participants were presented with five imaginary situations (e.g., “imagine that you burn your tongue on a hot drink”) and rated the pain

22

G. L. Wagener and A. Melzer

they would experience during these situations on a scale from 0 = no pain to 10 = worst possible pain. The PSQ had good internal consistency (α = .76). Empathy. To measure the relationship between empathy and pain perception, three of the four subscales of the German version of the Interpersonal Reactivity Index (IRI; [29, 30], namely empathic concern (e.g., “I often have tender, concerned feelings for people less fortunate than me”), perspective taking (e.g., “I sometimes try to understand my friends better by imagining how things look from their perspective”), and fantasy (e.g., “when I am reading an interesting story or novel, I imagine how I would feel if the events in the story were happening to me”). Each subscale was measured with four items on a 4-point-scale (1 = strongly disagree to 4 = strongly agree). The combined scale for trait-empathy had acceptable internal consistency (α = .67). Manipulation Check. To check if conditions were in fact perceived differently, participants rated two items on game content (e.g., “How brutal would you rate the game you just played”). Furthermore, they were asked about any difficulties with the game controls. Participants rated the items on a 5-point scale (1 = not at all to 5 = very much). Procedure. After participants gave their informed consent, they were alternately assigned by gender to one of two conditions (violent condition vs. non-violent condition). Afterwards, participants provided demographic data, rated the PANAS items, indicated their gaming habits, and rated the IRI items, followed by the Trait Aggression Questionnaire and the PSQ. Next, participants played one of two video games according to their experimental condition for 15 min. After gameplay, participants first rated the PANAS items again and answered the control items for their game perception. Finally, the CPT was performed. Participants were told that they had to hold their hand in the cold water as long as possible. Next, as a measure of aggressive behavior, participants had to indicate a time requirement in the CPT for the next participant as a measure of aggressive behavior. Finally, participants were remunerated, thanked for their participation, and debriefed. The entire experiment lasted for about 30 min.

2.2 Results All analyses were performed with IBM SPSS versions 25 and 27. The significance level was set at p < .05. Bonferroni correction for multiple comparisons was consistently applied in all three studies. Control Variables. In the CPT, there was no gender difference for pain perception, F(1, 60) = 1.49, p = .227, η2 p = .02. There were also no differences between conditions with regard to trait empathy, trait pain sensitivity, trait aggression, or gaming experience, ps ≥ .136. Personality traits did not correlate with pain perception, or pain tolerance, ps ≥ .118. Only VVG preference significantly correlated with trait empathy (r = .25, p = .041), trait aggression (r = .40, p = .001), and aggressive behavior (r = .32, p < .001). The manipulation check confirmed that the game conditions were perceived differently. Participants rated the violent game as more brutal (M = 4.27; SD = 1.01) than the non-violent game (M = 1.00; SD = 0.00), F(1, 64) = 347.53, p < .001, η2 = .84.

Comfortably Numb? Violent Video Games and Their Effects

23

Furthermore, VVG preference was correlated with change in positive affect (T2-T1) (r = .27, p = .031), pain tolerance (r = .28, p = .026), and proximal pain perception (r = -.28, p = .025). Hypotheses. A one-way ANOVA with condition as between-subjects variable (violent vs. non-violent condition) and proximal pain perception (pain ratings during CPT) as dependent variable tested H1. However, there was no significant effect, F(1, 62) < 0.01, p = .974, d < .01. Next, a one-way ANOVA with time in the CPT as dependent variable tested H3. Again, there was no significant effect, F(1, 64) = 1.97, p = .166, d = .35. Means and standard deviations are displayed in Table 1. To test H4, a one-way ANOVA was calculated with condition as between-subjects factor and time in the CPT allotted to the next participant as dependent variable, but did not reveal a significant effect, F(1, 64) = 0.04, p = .835, d = .05. Finally, two mixed-measures ANOVA tested H6. Again, condition served as between-subjects factor, and positive affect (PA score at T1 vs. PA score at T2) and negative affect (NA score at T1 vs. NA score at T2) served as within-subjects factors, respectively. Regarding positive affect, there was no significant within-subjects effect or between-subjects effect, p ≥ .10. However, the interaction between condition and PA score was significant, F(1, 60) = 23.11, p < .001, η2 p = .28. Contrasts revealed that for participants who played the violent game there was a significant decrease in positive affect from T1 to T2, F(1, 28) = 14.94, p = .001, η2 p = .35, see Fig. 1. In contrast, participants who played the non-violent game had a significant increase in positive affect, F(1, 32) = 7.31, p = .011, η2 p = .19. For the negative affect, there was a significant within-subjects effect, F(1, 62) = 13.80, p < .001, η2 p = .18, and a significant betweensubjects effect, F(1, 62) = 13.07, p = .001, η2 p = .17. However, the interaction between condition and the within-subjects factor of NA score was also significant, F(1, 62) = 24.45, p < .001, η2 p = .28. Separate analysis showed that the negative affect increased for participants who played the violent game, F(1, 30) = 22.62, p < .001, η2 p = .43. This was not the case for participants who played the non-violent game, F(1, 32) = 1.81, p = .188, η2 p = .05.

Fig. 1. Positive and negative affect before (T1) and after (T2) gameplay in the violent video game condition and the non-violent video game condition in study 1.

24

G. L. Wagener and A. Melzer

Table 1. Means and standard deviations for positive and negative affect at T1 and T2 as well as proximal pain perception and pain tolerance in the two game conditions in study 1. Measure

Violent condition M

Positive affect T1

Non-violent condition SD

3.06

0.60

M

SD 2.72

0.55

Positive affect T2

2.52

0.55

2.98

0.69

Negative affect T1

1.44

0.50

1.35

0.49

Negative affect T2

1.99

0.68

1.27

0.35

Proximal pain perception Pain tolerance (with exclusions) Pain tolerance (without exclusions) Aggressive Behavior (time allotted to next participant in CPT)

6.91

1.65

6.92

1.61

88.65

65.48

96.41

71.49

101.45

81.50

133.42

102.52

48.94

55.41

46.55

35.55

To test if gender influenced the effect of VVG exposure on pain tolerance, a two-way ANOVA was calculated with gender and condition (violent vs. non-violent) as betweensubjects factors and time in the CPT as dependent variable. Results did not show any significant main or interaction effects, Fs ≤ .82, ps ≥ .370, η2 p ≤ .02. Furthermore, an additional ANCOVA was run to test for the effect of condition (violent vs. non-violent condition) on pain tolerance (time in the CPT), including pain sensitivity (PSQ score) as covariate. The main effect of condition on pain tolerance did not reach the level of significance, F(1,54) = .02, p = .884, η2 p < .001. Pain sensitivity was also not a significant predictor of pain tolerance, F(1, 54) = 1.79, p = .187, η2 p = .03. Means and standard deviations are displayed in Table 2. Table 2. Means and standard deviations for pain tolerance, PSQ score, as well as Pain tolerance at the covariate mean level of PSQ Measure

Violent condition

Non-violent condition

M

SD

M

SD

Pain tolerance

90.40

65.85

96.41

71.49

Pain tolerance at PSQ = 3.88

91.98

12.49

94.65

13.17

3.94

1.65

3.59

1.40

PSQ score

2.3 Discussion As expected, there was a significant effect of VVG play on mood, decreasing positive affect and increasing negative affect. However, VVG play did not significantly affect pain

Comfortably Numb? Violent Video Games and Their Effects

25

tolerance in the CPT, proximal pain perception, or aggressive behavior. Can the absence of a desensitization effect be replicated, or was it just due to the use of inappropriate methods?

3 Study 2 Study 2 included different measures for pain and aggressive behavior. Pain tolerance and perception were assessed with the Medoc Pathway Pain and Sensory Evaluation System (see below). Heart rate variability (HRV) indicated the activity of the autonomic nervous system, with higher HRV relating to greater pain inhibition capacity [31]. The Competitive Reaction Time Task (CRTT) measured aggressive behavior. 3.1 Methods Participants. A total of 64 participants (M age = 27.57; SD = 11.57) took part in study 2 (50% females) at the University of Luxembourg. As in study 1, participants indicated how much they played video games (M = 2.22; SD = 1.23). They also rated their gaming experience (1 = not experienced, 2 = somewhat inexperienced, 3 = somewhat experienced, 4 = experienced). On average, participants were experienced on a medium level (M = 2.38; SD = 1.11). Again, participants rated the five items for violent video game preference (α = .91). Pain Tolerance and Proximal Pain Perception. To assess pain tolerance, pain stimuli were applied via a heat thermode. There were 9 trials, each lasting five seconds with an increase of 1 °C per trial. The first heat stimulus was set at 42 °C and the maximum heat stimulus was set at 50 °C. Between trials, the thermode immediately cooled to 32 °C. For each trial, participants rated their perceived pain on a 10-point numerical rating scale (0 = no pain and 10 = worst possible pain) [24]. If a participant rated a pain stimulus as 10, the task was immediately discontinued. Individual pain tolerance was the sum of the trials the participants had completed until they gave a maximum pain rating (10) or until all 9 trials were completed. Competitive Reaction Time Task. The CRTT is a flexible and powerful tool to assess the effect of aggression-eliciting stimuli [32]. For the task, participants were told that they would play a game against a pretend opponent who, unbeknownst to the participants, did not actually exist. In each of the nine rounds of the task, participants were told to press a key as quickly as possible when the green box on the monitor turned red. If the opponent was faster, the participants were exposed to a noise blast between 50 and 105 decibels through headphones. Prior to each round, participants indicated the amplitude and duration of the noise blast for their opponent if the opponent lost the round (aggressive behavior). The number of wins and defeats was determined in advance without the participants knowing. In the first round, the participants always lost. Heart Rate Variability. HRV was measured with two electrocardiography electrodes using a heart rate belt connected via Bluetooth to an iPad. One electrode was placed under the upper right clavicle and the other electrode was placed above the hip on the left side of

26

G. L. Wagener and A. Melzer

the body. Data was collected with the app HRV Logger and transferred to Artiifact [33] for further analyses. Here, the root mean square of successive differences (RMSSD) for the time domain and the absolute power of the high frequency-band (HF; 0.15–0.40 Hz) for the frequency domain were used [34]. RMSSD reflects beat-to-beat HRV and is robust against influential factors like respiration and is correlated to HF. RMSSD and HF are reliable measures for parasympathetic activity [34], which typically indicates the bodily functions when a person is at rest. Baseline HRV was assessed during the five minutes it took participants to answer questionnaire items, during gameplay, and during the pain perception task. Target Games. Participants played the same games as in study 1 for 15 min. Mood, Trait Aggression, Empathy, Pain Sensitivity. Violent Video Game Preference. Study 2 used the same measures as in study 1. Mood was again assessed at two time points with the PANAS scales [25]. The internal consistency before gameplay was good for positive affect (α = .85) and acceptable for negative affect (α = .67). For the post-measurement, the internal consistency was excellent both for positive affect (α = .91) and negative affect (α = .91). The German version of the Aggression Questionnaire [27] measured trait-aggression which showed very good internal consistency (α = .85). The combined scale for trait empathy that included the three subscales for emotional concern, fantasy, and perspective taking from the German version of the IRI [30] had good internal consistency (α = .82). With regard to participants’ inherent pain sensitivity, the PSQ-scale [28] showed an acceptable internal consistency (α = .67). Finally, the violent video game preference scale had excellent internal consistency (α = .91). Manipulation Check. Game perception was measured with four novel items (e.g., “How brutal would you rate the game you just played?”; α = .87). Another item tested if game mechanics or the effects of the controls of the game influenced participants (i.e., “How difficult was it for you to control the game?”). Items were rated on a 4-point scale (1 = not at all to 4 = very much). Procedure. After participants gave informed consent, they provided demographic information and rated the items of the questionnaire. In the meantime, baseline HRV was assessed. Then, participants were randomly assigned to one of the two conditions (violent condition vs. non-violent condition) and played for 15 min. During gameplay, the second HRV measurement was recorded. After gameplay, participants first rated the PANAS items again and filled in the control items for game perception. Then, the task for pain perception and tolerance was applied together with the third HRV measurement. Next, participants completed the CRTT task. Finally, they were remunerated, thanked for their participation, and debriefed. The entire study took about 45 min.

3.2 Results Control Variables. There were no significant group differences for trait aggression, empathy, pain sensitivity, or for the effect of game controls ps ≥ .102. In contrast, there

Comfortably Numb? Violent Video Games and Their Effects

27

was a significant difference between the two conditions on game perception, Welch’s F(1, 32.36) = 240.64, p < .001, η2 = .80. As expected, participants rated the VVG as significantly more brutal, more morally questionable, felt more guilt, and felt that they dealt out more pain in the game (M = 2.85; SD = 0.66) than the non-violent game (M = 1.02; SD = 0.10). In the CRTT, males (M = 5.75, SD = 1.92) behaved significantly more aggressively than females (M = 3.71, SD = 2.45), F(1, 62) = 13.71, p < .001, η2 = .18. There was also a gender effect for duration in the CRTT, with males (M = 5.05; SD = 2.19) choosing longer durations of noise blasts than females (M = 3.36, SD = 2.18), F(1, 62) = 9.59, p = .003, η2 = .13. However, no other gender effect was significant, ps ≥ .115. Aggressive behavior (CRTT intensity) was correlated with gaming experience, r = .35, p = .005, trait aggression, r = .29, p = .021, and pain perception, r = -.25, p = .043. Interestingly, trait empathy was positively related to gaming experience, r = .35, p = .005. Violent video game preference was positively correlated with intensity (r = .43, p < .001) and duration (r = .49, p < .001) in the CRTT only, but not to proximal pain perception, mood, HRV, or pain tolerance (ps ≥ .124). Hypotheses. To test H3, two one-way ANOVAs were calculated with game condition as independent variable, and pain perception (i.e., mean of pain ratings) and pain tolerance (i.e., sum of pain induction trials) as dependent variables, respectively. Game conditions did not differ for pain perception or pain tolerance, ps ≥ .520. Means and standard deviations are shown in Table 2. To test H4, a one-way ANOVA was calculated with condition as independent variable, intensity in the CRTT, and duration in the CRTT as dependent variables. There was no significant difference between conditions for CRTT intensity or CRTT duration, ps ≥ .620 (Table 3).2

Table 3. Means and standard deviations in the two game conditions for CRTT intensity, CRTT duration, pain perception, and pain tolerance in study 2. Measure

Violent condition Non-violent condition

Group Differences

M

SD

M

SD

Test

p

η2

CRTT Intensity

4.88

2.72

4.58

2.10

F = 0.25

.621

< .01

CRTT Duration

4.34

2.57

4.07

2.09

F = 0.22

.641

< .01

Pain Perception

5.28

1.82

5.00

1.58

F = 0.42

.518

.01

Pain Tolerance

8.22

1.69

8.44

0.91

F = 0.42

.520

.01

2 As noted by a reviewer, as there were significant gender differences in the CRTT gender should

be included as a covariate. Therefore, additional ANCOVAs were computed with gender as a covariate. Again, there were no significant effects of VVG exposure on aggressive behavior in the CRTT (p ≥ .771).

28

G. L. Wagener and A. Melzer

To test H5, a mixed ANOVA was calculated with condition as between-subjects variable, and RMSSD and HF as within-subjects variables. There was no significant between-subjects effect, F(1, 39) = 2.89, p = .097, η2 p = .07, but a significant withinsubjects effect for RMSSD, F(1.72, 66.99) = 8.92, p = .001, η2 p = .19. Mauchly’s test of sphericity was significant (Mauchly’s W = .84, p = .033) and therefore the Greenhouse-Geisser correction (ε = .86) was applied. Within-subjects contrasts confirmed a significant decrease in RMSSD from baseline (M = 44.18; SD = 26.41) to RMSSD during gameplay (M = 34.31; SD = 19.94), F(1, 39) = 7.82, p = .008, η2 p = .17, and from gameplay to pain perception (M = 45.57; SD = 21.81), F(1, 39) = 20.31, p < .001, η2 p = .34. More importantly, there was a significant interaction effect between condition and RMSSD change from gameplay to pain perception task. Only participants who played the VVG had a significant increase in RMSSD from gameplay to pain perception task, F(1, 39) = 5.24, p = .028, η2 p = .12 (see Fig. 2). For HF, there was also a significant within-subjects effect for HF, F(2, 78) = 7.15, p = .001, η2 p = .16. During gameplay (M = 0.04, SD = 0.02), there was a significant decrease in HF compared to baseline (M = 0.04, SD = 0.03), F(1, 39) = 4.45, p = .041, η2 p = .10, whereas HF significantly increased from gameplay to the pain perception task (M = 0.05; SD = 0.03), F(1, 39) = 15.62, p < .001, η2 p = .29 (see Fig. 3). There was no significant interaction effect, F(2, 78) = 1.36, p = .262, η2 p = .03 or between-subjects effect, F(1, 39) = 2.25, p = .142, η2 p = .06.

Fig. 2. Mean RMSSD in the VVG condition and the non-violent video game condition at baseline (T1), during gameplay (T2) and during CRTT (T3) in study 2.

Finally, two mixed-measures ANOVA tested H6. Condition served as betweensubjects factor, and positive affect (PA score at T1 vs. PA score at T2) and negative affect (NA score at T1 vs. NA score at T2) were the within-subject factors, respectively. For positive affect, only the interaction between condition and PA score was significant, F(1, 62) = 4.17, p = .045, η2 p = .06, but separate analyses did not reveal any significant effects, ps ≥ .130. For negative affect, there was a significant within-subjects effect, F(1, 62) = 5.50, p = .022, η2 p = .08, indicating a significant increase in negative affect from

Comfortably Numb? Violent Video Games and Their Effects

29

Fig. 3. Mean HF in the VVG condition and the non-violent video game condition at baseline (T1), during gameplay (T2) and during CRTT (T3) in study 2.

T1 (M = 1.32; SD = 0.33) to T2 (M = 1.48; SD = .63), and a significant betweensubjects effect, F(1, 62) = 14.33, p < .001, η2 p = .19. More importantly, the interaction effect between condition and the within-subjects factor of NA score was significant, F(1, 62) = 10.58, p = .002, η2 p = .15. Separate analysis showed that participants who played the violent game showed a significant increase in negative affect from T1 (M = 1.32; SD = 0.33) to T2 (M = 1.76; SD = 0.78.), F(1, 31) = 8.65, p = .006, η2 p = .22, see Fig. 4. In the non-violent game condition, there was no significant within-subjects effect for NA score from T1 (M = 1.27; SD = .27) to T2 (M = 1.20; SD = 0.20), F(1, 31) = 2.19, p = .149, η2 p = .07.

Fig. 4. Positive affect and negative affect in the VVG condition and the non-violent video game condition at the beginning of the experiment (T1) and after gameplay (T2) in study 2.

30

G. L. Wagener and A. Melzer

3.3 Discussion Study 2 confirmed the results of study 1, using different outcome measures. There were no significant effects of VVG play on aggressive behavior, pain perception, or on pain tolerance. Again, VVP play lead to increased negative effect. In study 2, however, VVG play had no influence on positive affect. Since the level of immersion in the game mediates the effects of VVG [35], study 3 was conducted to replicate the results of the first two studies and test the level of immersion as an additional factor.

4 Study 3 The level of immersion in the game mediates the effects of VVG [35] and influences pain sensitivity [21]. Therefore, participants played either a violent-immersive, a nonviolent-immersive or a non-violent-non-immersive video game in study 3. 4.1 Methods Participants. Study 3 involved 75 participants (M age = 27; SD = 10.06; 42% females). Participants indicated to play VG on a medium level (M = 2.22; SD = 1.23). Pain Perception. To measure distal pain perception, participants rated the level of pain they perceived in pictures. The pictures were pilot tested with N = 50 student participants. Each of the 15 neutral pictures and 17 pain-related pictures of the IAPS (International Affective Picture System; [36]) were rated on a numerical pain rating scale from 0 (=no pain) to 10 (=worst possible pain) [24]. From these pictures, 10 neutral (α = .76) and 10 pain-related pictures (α = .92) with the highest reliabilities were chosen. A dependent sample t-test showed that the pain-related pictures were rated as significantly more painful (M = 7.11, SD = 1.89) than the neutral pictures (M = 0.30, SD = 0.52), t(49) = 26.01, p < .001, d = 4.91. In study 3, participants rated these 20 items in randomized order on the same numerical pain rating scale used in the pilot study. Proximal pain perception was again measured with the CPT (see study 1). Heart Rate Variability. Again, there were three heart rate measurement times during the study. Data from T1 served as a baseline, data from T2 were collected during gameplay, and data from T3 were recorded during the CPT. Due to technical difficulties, T1 data for four participants could not be used. Mood, Trait Aggression, Pain Sensitivity. Study 3 used the same measures as studies 1 and 2. Mood was again assessed at two timepoints with the PANAS [25]. In study 3, the internal consistency at t1 was very good for positive affect (α = .86) and acceptable for negative affect (α = .63). For t2, the internal consistency ranged from very good to excellent for the positive (α = .92) and negative affect (α = .87). The two subscales for anger and hostility from the German version of the Aggression Questionnaire [34] were combined to general trait aggression, which showed very good internal consistency (α = .85). The PSQ was again used to measure participants’ inherent pain sensi-tivity [35] and had acceptable internal consistency (α = .69).

Comfortably Numb? Violent Video Games and Their Effects

31

Target Games. In the violent condition, Sniper Elite III for the PlayStation4 was used due to its high violence ratings (age rating label “PEGI 18”) and high level of immersion. In Sniper Elite III, the player shoots down enemy WW2 soldiers as a sniper. The game contains explicit graphical representations of extreme violence and blood. In the nonviolent immersive condition, the racing game Mario Kart 8 was played on the Wii U console. In the non-violent non-immersive condition, Yoshi’s Fruit Cart for the Wii U was chosen as it is neither violent nor immersive. The game required participants to draw line paths with the Stylus so that the game character Yoshi can collect points (i.e., fruits). All the three games had simple controls that were easy to explain to participants. Empathy. Trait empathy was measured using the two subscales for emotional concern and perspective taking from the German version of the IRI [30]. The combined scale had good internal consistency (α = .73). In addition to the IRI, media-based empathy (MBE; [37]) was used to measure the ability of participants to feel empathy for fictitious characters (4 items; e.g., “Media reports about what is happening in the world are very close to me.”), as well as their ability to immerse themselves in VG (5 items; e.g., “I experience very strong feelings when I play good video games.”). The items were rated on a five-point Likert scale (1 = I strongly disagree to 5 = I strongly agree; α = .83). Manipulation Check. Game perception was measured with five items on competition, frustration, time pressure, brutality, and immersion. The items (e.g., “how brutal would you rate the game you just played?”) were rated on a 5-point scale (1 = not at all to 5 = very much). The scale showed good internal consistency (α = .74). Procedure. After participants gave informed consent, the electrodes for HRV assessment were applied. Then, questions on demographics, trait empathy, trait pain sensitivity were answered, and the PANAS items rated. Next, participants were randomly assigned to one of the three game conditions. After familiarizing themselves with the game controls, they played for 15 min. Then, participants rated the PANAS items again and the items for trait aggression. Next, the CPT was administered. Then, the pictorial stimuli were shown to assess distal pain perception. Finally, participants were remunerated, thanked and debriefed. The experiment took about 45 min.

4.2 Results Control Variables. There were no significant differences between conditions for trait empathy, trait aggression, media-based empathy, or pain sensitivity, ps ≥ .37. Also, there were no differences between the genders, ps ≥ .40. Hypotheses. The between-subjects ANOVA that tested the differences between the three game conditions for proximal pain perception (H1) showed no significant effect, F(2, 72) = 0.25, p = .783, η2 p = .01. Neither the ANOVA on distal pain perception (H2) nor the comparison of conditions in terms of pain tolerance (H3) revealed significant effects, p ≥ .433. Means and standard deviations are displayed in Table 4.

32

G. L. Wagener and A. Melzer

Table 4. Means and standard deviations in the different game conditions for proximal pain perception, distal pain perception, and pain tolerance in study 3. Measure

Violent condition

Non-violent-immersive condition

Non-violent-non-immersive condition

M

M

M

SD

SD

SD

Proximal Pain Perception

6.48

1.12

6.30

2.21

6.63

1.42

Distal Pain Perception

7.87

1.74

8.10

0.99

7.81

1.24

Pain Tolerance 91.80

58.22

100.60

62.62

78.20

63.13

Next, two repeated-measures ANOVAs tested if VVG exposure influenced mood (H6). Condition served as between-subjects factor, whereas positive affect (PA score at T1 vs. PA score at T2) and negative affect (NA score at T1 vs. NA score at T2) were the within-subject factors, respectively. No significant effects were found, p ≥ .085. However, separate analyses showed that there was a significant decrease in positive mood for participants who played the violent game, F(1, 23) = 8.87, p = .007, η2 p = .28, see Fig. 5. There was no significant difference in mood after playing the nonviolent-immersive or the non-violent-non-immersive video game, Fs ≤ .33, ps ≥ .592, η2 p ≤ .01. For negative affect, there was only a significant interaction effect, F(2, 71) = 6.79, p = .002, η2 p = .16. Separate analysis showed a significant increase in negative affect for participants who had played the violent game, F(1, 23) = 12.51, p = .002, η2 p = .35, which was not the case in the other conditions, Fs ≤ 1.39, p ≥ .249, η2 p ≤ .06. Means and standard deviations are displayed in Table 5. Two additional ANCOVAs with pain sensitivity (PSQ score) as covariate were calculated. There were no significant effects on pain tolerance (time in the CPT) or proximal pain perception, p ≥ .15. Pain sensitivity did not predict pain tolerance, F(26, 26.70), p = .970, η2 p = .49, but proximal pain perception, F(26, 29.97) = 2.58, p = .007, η2 p = .691. To test H5, a mixed ANOVA was calculated with RMSSD and with HF as withinsubjects factors and condition as between-subjects factor. For RMSSD, Mauchly’s test of sphericity was significant (p < .001) and so the Greenhouse-Geisser correction was applied (ε = .83). The analysis revealed a significant within-subjects effect for RMSSD, F(1.66, 119.49) = 5.19, p = .011, η2 p = .07, with a significant increase in RMSSD from gameplay (M = 39.17; SD = 20.78) to when participants did the CPT (M = 43.70; SD = 22.83), F(1, 72) = 5.66, p = .020, η2 p = .07. However, there was no significant between-subjects effect or interaction effect, p ≥ .430 (see Fig. 6). For HF, Mauchly’s test of sphericity was significant (p < .001) and so the Greenhouse-Geisser correction was applied (ε = .77). The analysis revealed no significant effects p ≥..083.

Comfortably Numb? Violent Video Games and Their Effects

33

Fig. 5. Positive affect and negative affect in the three game conditions at the beginning of the experiment (T1) and after gameplay (T2) in study 3.

Table 5. Means and standard deviations for positive and negative affect at T1 and T2 across the three conditions in study 3. Measure

Violent condition

Non-violent-immersive condition

Non-violent-non-immersive condition

M

SD

M

SD

M

SD

Positive affect T1

2.27

0.58

2.12

0.67

1.96

0.55

Positive affect T2

1.81

0.75

2.20

0.82

1.84

0.93

Negative affect T1

0.26

0.28

0.31

0.28

0.34

0.27

Negative affect T2

0.66

0.68

0.30

0.50

0.26

0.33

4.3 Discussion Again, VVG play significantly affected mood, leading to decreased positive affect and increased negative affect. However, like study 1 and 2, there were no significant effects of VVG play on pain-related indicators. In addition, level of immersion had no effect on the results for the outcome variables.

34

G. L. Wagener and A. Melzer

Fig. 6. Mean RMSSD at baseline, during gameplay, and during the CPT across the three conditions in study 3.

5 General Discussion Does playing VVG increase aggression while numbing and desensitizing pain? Three lab experiments tested the effects of VVG on pain-related indicators and aggressive behavior. The empirical results of the present studies are clear: there were no significant effects of direct VVG exposure on behavioral data of aggressive behavior, pain tolerance, or pain perception. In contrast, the VVG preference of participants was associated with higher pain tolerance and lower proximal pain perception in study 1, and VVG preference was associated with higher trait aggression and higher aggressive behavior in study 2. We may speculate that people with higher trait aggression and those who behave more aggressively are drawn more toward VVG and in-game depictions of violence. Notwithstanding this, at least the present findings suggest that VVG exposure itself does not lead to an increase in aggressive behavior or to desensitization. All three studies confirmed the negative effect that VVG play has on mood. Compared to a nonviolent VG participants in the VVG condition felt significantly worse after playing—their positive affect decreased, while negative affect increased. This finding is in line with prior research that showed VVG effects on mood [15, 38]. Pain is a complex and multidimensional construct, modulated by individual, physiological, and contextual factors [2]. Based on the findings from the present studies, it appears that physiological factors as well as personality traits dominate pain responses. Perhaps the gaming episode in the present studies was too weak as contextual factor to show any effect on pain perception or pain tolerance. In contrast, individual factors affected pain tolerance, pain perception, and aggressive behavior. Significant gender differences were found on aggressive behavior in study 1, with males being more aggressive than females. Personality traits (i.e., trait aggression, trait pain sensitivity, and violent video game preference) were related to pain tolerance, pain perception, and positive affect (study 1), and to aggressive behavior (study 2). Although some authors assume that situational factors influence aggression more than personality factors [39], this was not the case in the present studies. Similar to the presented results, other studies also

Comfortably Numb? Violent Video Games and Their Effects

35

show that personality factors rather than game violence are associated with aggression [40]. Markey and Markey concluded that personality factors and personal predispositions moderate negative VVG effects, thus making some individuals more vulnerable to VVG effects than others [41]. In addition, personality factors are related to both VVG preferences and VVG exposure [42, 43] and can moderate physiological reactions to VVG exposure [44]. Thus, when confronted with VG violence, individuals without these predispositions should not show an increase in aggression or desensitization. In study 2, HRV (RMSSD and HF) significantly decreased from baseline to gameplay and increased again to during the CRTT across both conditions. In study 3, HRV (RMSSD) significantly increased from gameplay to the CPT for participants across all gaming conditions. This shows that participants experienced decreased parasympathetic activity during gameplay, but higher parasympathetic activity after gameplay, possibly representing greater tension during and a reduction in tension after gameplay. This is in line with previous research showing an increase in parasympathetic activity after VVG play [45]. However, there were no significant differences between participants playing a violent or a non-violent game. Furthermore, research has shown that parasympathetic activity is decreased during the experience of pain [31]. This was also not the case in the present study as RMSSD increased during the CPT for participants in study 3. It is unclear why playing a VG decreased parasympathetic activity in study 2, and if HRV increased during the CRTT due to the prior gaming effects or due to the nature of the CRTT or other variables. Future studies should incorporate explicit measures that in addition to objective physiological measures to disentangle these findings, assess the perceived physiological states of participants. Playing a VG can indeed lead to a decrease in pain perception [21, 22, 46, 47]. It is possible that the assumed distractive, hypoalgesic effects of playing VG is independent of violent content. In the presented studies, however, there was no control condition in which participants did not play a VG. Therefore, it cannot be ruled out that playing VG generally leads to desensitization to pain. Based on the present findings, we can only conclude that the content of the game, violent or not, has no pain-related effects. Regarding limitations, the large variability in the present student-dominated samples with only a few VG enthusiasts may have contributed to the lack of pain-related effects detected. Sample sizes with few “hardcore gamers” might not be sufficient to detect small effects that would be expected for the desensitization effects of VVG [10]. Another limitation of the present studies lies in their design. The games used in the present studies varied greatly regarding content, game design, graphical appearance, and game controls. Future studies should limit these potentially influential factors by using more comparable target games, or ideally, by using passages (nonviolent vs. violent) from the same game. Moreover, although some authors have argued that even shortterm exposure to VVG can lead to desensitization [17], playing a violent game for only 15 min might be too short to influence later aggression, pain perception, or pain tolerance measures. Future studies should therefore increase playing time to increase the generalizability and reliability of findings. In summary, the present results are consistent with findings that exposure to VG violence does not have a desensitization effect on aggression [48, 49], pain perception

36

G. L. Wagener and A. Melzer

[50, 51], or empathic response [52]. However, the effects of violent games may be moderated by specific personality factors not tested here that make people more susceptible to the effects, increasing the likelihood of desensitization, aggression and pain tolerance, while at the same time decreasing pain perception and empathic concern. Future studies should therefore take a closer look into the moderating role of personality factors on the effects of (violent) video games on the players.

References 1. Calvert, S.L., et al.: The American psychological association task force assessment of violent video games: science in the service of public interest. Am. Psychol. 72, 126–143 (2017). https://doi.org/10.1037/a0040413 2. Melzack, R.: Evolution of the neuromatrix theory of pain. In: The Prithvi Raj Lecture: Presented at the Third World Congress of World Institute of Pain, Barcelona 2004. Pain Practice. 5, pp. 85–94 (2005). https://doi.org/10.1111/j.1533-2500.2005.05203.x 3. Bushman, B.J., Huesmann, L.R.: Twenty-five years of research on violence in digital games and aggression revisited: a reply to. Eur. Psychol. 19, 47–55 (2014). https://doi.org/10.1027/ 1016-9040/a000164 4. Bushman, B.J., Gollwitzer, M., Cruz, C.: There is broad consensus: media researchers agree that violent media increase aggression in children, and pediatricians and parents concur. Psychol. Pop. Media Cult. 4, 200–214 (2015). https://doi.org/10.1037/ppm0000046 5. Ferguson, C.J.: Pay no attention to that data behind the curtain: on angry birds, happy children, scholarly squabbles, publication bias, and why betas rule metas. Perspect Psychol Sci. 10, 683–691 (2015). https://doi.org/10.1177/1745691615593353 6. Ferguson, C.J.: Aggressive video games research emerges from its replication crisis (Sort of). Current Opinion in Psychology 36, 16 (2020) 7. Ferguson, C.J., Colwell, J.: Understanding why scholars hold different views on the influences of video games on public health: opinions on video game influences on public health. J Commun. 67, 305–327 (2017). https://doi.org/10.1111/jcom.12293 8. Scharrer, E., Kamau, G., Warren, S., Zhang, C.: Violent video games DO promote aggression. 28 9. Allen, J.J., Anderson, C.A., Bushman, B.J.: The general aggression model. Curr. Opin. Psychol. 19, 75–80 (2018). https://doi.org/10.1016/j.copsyc.2017.03.034 10. Anderson, C.A., et al.: Violent video game effects on aggression, empathy, and prosocial behavior in Eastern and Western countries: a meta-analytic review. Psychol. Bull. 136, 151– 173 (2010). https://doi.org/10.1037/a0018251 11. Bartholow, B.D., Bushman, B.J., Sestir, M.A.: Chronic violent video game exposure and desensitization to violence: behavioral and event-related brain potential data. J. Exp. Soc. Psychol. 42, 532–539 (2006). https://doi.org/10.1016/j.jesp.2005.08.006 12. Engelhardt, C.R., Bartholow, B.D., Kerr, G.T., Bushman, B.J.: This is your brain on violent video games: neural desensitization to violence predicts increased aggression following violent video game exposure. J. Exp. Soc. Psychol. 47, 1033–1036 (2011). https://doi.org/10. 1016/j.jesp.2011.03.027 13. Gentile, D.A., Swing, E.L., Anderson, C.A., Rinker, D., Thomas, K.M.: Differential neural recruitment during violent video game play in violent- and nonviolent-game players. Psychol. Pop. Media Cult. 5, 39–51 (2016). https://doi.org/10.1037/ppm0000009 14. Hartmann, T., Vorderer, P.: It’s okay to shoot a character: moral disengagement in violent video games. J. Commun. 60, 94–119 (2010). https://doi.org/10.1111/j.1460-2466.2009.014 59.x

Comfortably Numb? Violent Video Games and Their Effects

37

15. Lee, E.-H., et al.: The effects of video games on aggression, sociality, and affect: a metaanalytic study. KOSES. 23(4), 41–60 (2020). https://doi.org/10.14695/KJSOS.2020.23.4.41 16. Mathiak, K.A., Klasen, M., Weber, R., Ackermann, H., Shergill, S.S., Mathiak, K.: Reward system and temporal pole contributions to affective evaluation during a first person shooter video game. BMC Neurosci. 12, 66 (2011). https://doi.org/10.1186/1471-2202-12-66 17. Bushman, B.J., Anderson, C.A.: Comfortably numb: desensitizing effects of violent media on helping others. Psychol Sci. 20, 273–277 (2009). https://doi.org/10.1111/j.1467-9280.2009. 02287.x 18. Funk, J.B., Baldacci, H.B., Pasold, T., Baumgardner, J.: Violence exposure in real-life, video games, television, movies, and the internet: is there desensitization? J. Adolesc. 27, 23–39 (2004). https://doi.org/10.1016/j.adolescence.2003.10.005 19. Brockmyer, J.F.: Desensitization and violent video games. Child Adolesc. Psychiatr. Clin. N. Am. 31, 121–132 (2022). https://doi.org/10.1016/j.chc.2021.06.005 20. Primack, B.A., et al.: Role of video games in improving health-related outcomes. Am. J. Prev. Med. 42, 630–638 (2012). https://doi.org/10.1016/j.amepre.2012.02.023 21. Weger, U.W., Loughnan, S.: Virtually numbed: Immersive video gaming alters real-life experience. Psychon. Bull. Rev. 21(2), 562–565 (2013). https://doi.org/10.3758/s13423-0130512-2 22. Stephens, R., Allsop, C.: Effect of manipulated state aggression on pain tolerance. Psychol Rep. 111, 311–321 (2012). https://doi.org/10.2466/16.02.20.PR0.111.4.311-321 23. Miedzobrodzka, E., van Hooff, J.C., Konijn, E.A., Krabbendam, L.: Is it painful? playing violent video games affects brain responses to painful pictures: an event-related potential study. Psychology of Popular Media. 11, 13–23 (2022). https://doi.org/10.1037/ppm0000290 24. Karcioglu, O., Topacoglu, H., Dikme, O., Dikme, O.: A systematic review of the pain scales in adults: which to use? Am. J. Emerg. Med. 36, 707–714 (2018). https://doi.org/10.1016/j. ajem.2018.01.008 25. Watson, D., Anna, L., Tellegen, A.: Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales. 8 26. Buss, A.H., Perry, M.: The Aggression Questionnaire. 8 27. Herzberg, P.Y.: Faktorstruktur, Gütekriterien und Konstruktvalidität der deutschen Übersetzung des Aggressionsfragebogens von Buss und Perry. Zeitschrift für Differentielle und Diagnostische Psychologie. 24, 311–323 (2003). https://doi.org/10.1024/0170-1789.24.4.311 28. Ruscheweyh, R., Marziniak, M., Stumpenhorst, F., Reinholz, J., Knecht, S.: Pain sensitivity can be assessed by self-rating: Development and validation of the Pain Sensitivity Questionnaire. Pain 146(1-2), 6574 (2009) 29. Davis, M.H.: Measuring individual differences in empathy: evidence for a multidimensional approach. J. Pers. Soc. Psychol. 44, 113–126 (1983). https://doi.org/10.1037/0022-3514.44. 1.113 30. Paulus: (2009) 31. Forte, G., Troisi, G., Pazzaglia, M., Pascalis, V.D., Casagrande, M.: Heart rate variability and pain: a systematic review. Brain Sci. 12, 153 (2022). https://doi.org/10.3390/brainsci1 2020153 32. Warburton, W.A., Bushman, B.J.: The competitive reaction time task: the development and scientific utility of a flexible laboratory aggression paradigm. Aggr Behav. 45, 389–396 (2019). https://doi.org/10.1002/ab.21829 33. Kaufmann, T., Sütterlin, S., Schulz, S.M., Vögele, C.: ARTiiFACT: a tool for heart rate artifact processing and heart rate variability analysis. Behav Res. 43, 1161–1170 (2011). https://doi. org/10.3758/s13428-011-0107-7 34. Shaffer, F., Ginsberg, J.P.: An overview of heart rate variability metrics and norms. Front. Public Health. 5, 258 (2017). https://doi.org/10.3389/fpubh.2017.00258

38

G. L. Wagener and A. Melzer

35. Lull, R.B., Bushman, B.J.: Immersed in violence: presence mediates the effect of 3D violent video gameplay on angry feelings. Psychol. Pop. Media Cult. 5, 133–144 (2016). https://doi. org/10.1037/ppm0000062 36. Lang, P.J., Bradley, M.M., Cuthbert, B.N.: International affective picture system (IAPS): affective ratings of pictures and instruction manual. Technical report A-8. University of Florida, Gainesville, FL (2008) 37. Happ, C., Pfetsch, J.: Medienbasierte Empathie (MBE): Entwicklung eines Instruments zur Erfassung empathischer Reaktionen bei Mediennutzung. Diagnostica 62, 110–125 (2016). https://doi.org/10.1026/0012-1924/a000152 38. Saleem, M., Anderson, C.A., Gentile, D.A.: Effects of prosocial, neutral, and violent video games on college students’ affect: violent video games and students’ affect. Aggr Behav. 38, 263–271 (2012). https://doi.org/10.1002/ab.21427 39. Hasan, Y., Eldous, H.: The role of personality traits and situational factors as determinants of aggression. TOPSYJ. 13, 282–288 (2020). https://doi.org/10.2174/1874350102013010282 40. Winkel, M., Novak, D.M., Hopson, H.: Personality factors, subject gender, and the effects of aggressive video games on aggression in adolescents. J. Res. Pers. 21, 211–223 (1987). https://doi.org/10.1016/0092-6566(87)90008-0 41. Markey, P.M., Markey, C.N.: Vulnerability to violent video games: a review and integration of personality research. Rev. Gen. Psychol. 14, 82–91 (2010). https://doi.org/10.1037/a0019000 42. Greitemeyer, T.: Everyday sadism predicts violent video game preferences. Personality Individ. Differ. 75, 19–23 (2015). https://doi.org/10.1016/j.paid.2014.10.049 43. Greitemeyer, T., Sagioglou, C.: The longitudinal relationship between everyday sadism and the amount of violent video game play. Personality Individ. Differ. 104, 238–242 (2017). https://doi.org/10.1016/j.paid.2016.08.021 44. Wagener, G.L.: Presented at the (2020) 45. Ivarsson, M., Anderson, M., Åkerstedt, T., Lindblad, F.: Playing a violent television game affects heart rate variability. Acta Paediatr. 98, 166–172 (2009). https://doi.org/10.1111/j. 1651-2227.2008.01096.x 46. Barcatta, K., Holl, E., Battistutta, L., van der Meulen, M., Rischer, K.M.: When less is more: investigating factors influencing the distraction effect of virtual reality from pain. Front. Pain Res. 2, 800258 (2022). https://doi.org/10.3389/fpain.2021.800258 47. Gupta, A., Scott, K., Dukewich, M.: Innovative technology using virtual reality in the treatment of pain: does it reduce pain via distraction, or is there more to it? Pain Med. 19, 151–159 (2018). https://doi.org/10.1093/pm/pnx109 48. Goodson, S., Turner, K.J., Pearson, S.L., Carter, P.: Violent video games and the P300: no evidence to support the neural desensitization hypothesis. Cyberpsychol. Behav. Soc. Netw. 24, 48–55 (2021). https://doi.org/10.1089/cyber.2020.0029 49. Read, G.L., Ballard, M., Emery, L.J., Bazzini, D.G.: Examining desensitization using facial electromyography: violent videogames, gender, and affective responding. Comput. Hum. Behav. 62, 201–211 (2016). https://doi.org/10.1016/j.chb.2016.03.074 50. Gao, X., Pan, W., Li, C., Weng, L., Yao, M., Chen, A.: Long-time exposure to violent video games does not show desensitization on empathy for pain: an fmri study. Front. Psychol. 8, 650 (2017). https://doi.org/10.3389/fpsyg.2017.00650 51. Kühn, S., Kugler, D., Schmalen, K., Weichenberger, M., Witt, C., Gallinat, J.: The myth of blunted gamers: no evidence for desensitization in empathy for pain after a violent video game intervention in a longitudinal fMRI study on non-gamers. Neurosignals 26, 22–30 (2018). https://doi.org/10.1159/000487217 52. Szycik, G.R., Mohammadi, B., Münte, T.F., te Wildt, B.T.: Lack of Evidence That neural empathic responses are blunted in excessive users of violent video games: an fMRI study. Front. Psychol. 8, 174 (2017). https://doi.org/10.3389/fpsyg.2017.00174

Discovering the Motivational Constitution of ‘Playing Games for Fun’ Kai Tuuri1(B)

and Jukka Vahlo1,2

1

2

Department of Music, Art and Culture Studies, University of Jyv¨ askyl¨ a, P.O. Box 35, 40014 Jyv¨ askyl¨ a, Finland [email protected] School of Economics, Centre for Collaborative Research, University of Turku, 20014 Turku, Finland [email protected]

Abstract. Regardless of its all-encompassing and ubiquitous nature, game and play researchers have often steered away from applying fun as a research concept. If a concept seems to be associated with everything, it logically follows that the concept lacks explanatory power. In this paper, we do not merely settle for the blunt conclusion that fun is not an interesting research concept. Rather we start to explore the phenomenon of fun by approaching it through three lenses: motivation to play, gameplay experience, and psychological need satisfaction. By analyzing two large survey samples collected in Finland (N = 879) and South-Korea (N = 1519), we cluster survey participants into player types according to their gameplay motivations. It is revealed that all players are more motivated by fun than by other need-based gaming motives, but also that a significant minority of players are only motivated by fun. By studying player preferences of the player types, it is furthermore highlighted that these Fun-Seekers generally dislike most gameplay activities and differ from other player types also regarding their genre play habits. Practical and theoretical implications of these findings are discussed.

Keywords: Fun

1

· Gameplay · Motivation · Self-determination

Introduction

Video games are fun. However oversimplified this argument may be, it gets continuously repeated both in the ongoing cultural discourse on video games as well as in industrial and academic accounts of them. It is very commonplace to just state that “I play games just for fun” or that “games are nothing but good fun”. Just to give an example of the importance of fun in the academic discussion, in This work has been funded by Business Finland (projects 9214/31/2019 and 864/31/2016), Academy of Finland (Centre of Excellence in Game Culture Studies, project n:o 312397), and supported by Kinrate Analytics Ltd, a private company specialized in player market analytics. c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 39–46, 2022. https://doi.org/10.1007/978-3-031-20212-4_3

40

K. Tuuri and J. Vahlo

his book about game-based learning, Marc Prensky [6] has listed twelve characteristics of video games that contribute to their ability to be engaging. The very first of them states that games are essentially a form of fun. However, as an academic concept, “fun” is highly ambiguous in its meaning, and the understanding of the concept rarely gets elaborated in papers discussing or utilizing this concept in research [14]. Our focus in this work-in-progress paper is on shedding more light on fun as a reason to play games. While also acknowledging how the concept might refer to certain types of games and features of gameplay, we are here primarily interested in describing fun in motivational terms. Instead of settling for a singular concept of fun, our intention is to discover the motivational constitution of this concept. That is, what kind of motives to play games and motivational factors of gameplay are associated with fun. Already at the dawn of video game research, Chris Crawford [2] has discussed the potential differences between asking “why do people play games in general?” and “what makes a game more fun than another?” According to Crawford, the key difference lies between the motivating and enjoyment factors of games. Admitting an inevitable dependence between the two, motivating factors refer to the reasons for approaching games (even the ones without much enjoyment but, e.g., a good exercise) while enjoyment factors help people to choose between games on the basis of fun, that is, the gameplay and sensory based gratification it offers. However, in order to be fun, a game must sufficiently satisfy the approach motivations of the player [2]. On the basis of such a formulation, fun clearly is a criterion for game choice and for continued play, but it is not necessarily to be seen as a genuine approach motivation to play games. This formulation, however, leaves much room for clarifying the relationship between motivation and fun, especially in regards to studying what kind of things in gameplay constitute fun in terms of motivational approach towards games. The terms “fun” and “enjoyment” are frequently considered synonyms in the literature [14]. In particular, enjoyment and pleasure emphasize positive connotations of fun. But such a meaning should not be restricted to superficial aspects of the concept, as even the most serious things in life (especially the ones we are passionate about) give us enjoyment and make us return to the corresponding activities [6]. On the other hand, “fun” can also promote negative connotations, for example, when understood in more frivolous terms of amusement or ridicule [6]. However, one should note that in regard to play, pleasures are merely not submissive to (superficial) fun and happiness, since engagement with play opens up a whole spectrum of variations of pleasure in the world [10]. How should we then conceptualize these engaging things in life that promote substantial pleasure and enjoyment for people? We may look for answers to this question in Ryan and Deci’s [7] influential Self-Determination Theory (SDT). It is a macrotheory of human motivation and organismic growth, with a focus on curiosity and explorative behavior as a manifestation of intrinsic motivation. According to the SDT [7], three basic psychological needs (autonomy, competence and relatedness) energize motivational processes and promote harmonious development of the human organism within the world. Satisfaction of the basic need for actualizing one’s autonomic self is considered to be the primary element

Discovering the Motivational Constitution of ‘Playing Games for Fun’

41

of this theory, essentially referring to activities that embrace one’s spontaneous curiosity, interest and free will. This seems to align well with notions that the disposition of having “fun” is essentially a voluntary (intrinsically motivated) attitude of perceiving situations as enjoyable (e.g., [1]), as well as with game research literature emphasizing players’ participation in game experiences as being characteristically voluntary and autonomous (e.g., [4,5,12]). The models of game motivation research are also typically built upon the premise of an autonomous player (e.g., [9,15]). In addition to autonomy, the basic needs of competence and relatedness also provide potential approaches. For example, as the flow-theory [3] proposes, the balance between a player’s skills and a game’s challenges should arguably result in pleasurable experiences relating to one’s competence. Tamborini et al. [13] have established that enjoyment indeed derives from the satisfaction of SDT needs, which however, do not necessarily include a purely pleasure-driven search for enjoyment.

Fig. 1. The circle of fun, incorporating the perspectives of Motivation, Gameplay experience and Self-Determination.

In this study we conceptualize the SDT basic needs satisfaction as the self-reflective ways how fun is in general constituted in personal life. Selfdetermination thus provides an approach for scrutinizing both the varieties of pleasure afforded by gameplay and fun as a motivational disposition (see Fig. 1). Furthermore, this approach opens up a possibility to utilize three “lenses” in conceptualizing fun in terms of (1) motivation, (2) situated gameplay experience, and (3) reflective constitution of the self. On the basis of two survey datasets, we will next report a brief empirical investigation of how “fun” appears as a factor of gameplay motives and preferences through these three lenses.

2

Methods

Two large survey samples were collected within the ongoing project from Finland (N = 879) and from South-Korea (N = 1519) in cooperation with two

42

K. Tuuri and J. Vahlo

market research companies. Both of the data collection processes were designed to produce samples that would represent the age (ages from 18 to 65) and gender distribution of the said countries. Prior to making any analyses, a total of 52 survey responses were removed from the Finnish sample and 131 from the Korean sample as these respondents were identified as outliers. After the data cleaning process, the Finnish sample consisted of 827 respondents (49.0% male, 51.0% female, mean age 41.4) and the Korean sample of 1,388 respondents (50.4% male, 49.6% female, mean age 39.1). Both of the surveys included the Intrinsic Motivations to Gameplay (IMG) inventory, the Gameplay Activity inventory (GAIN), a 17-item inventory on video game genre play, a five-item inventory on preferred game modes (e.g., single-player computer games, multiplayer mobile games), and demographic questions. The 15-IMG is a SDT compatible five-factor inventory on motives to play digital games [15]. It assesses general gaming motives of Autonomy (α = 0.92), Competence (α = 0.89), Relatedness (α = 0.89), Immersion (α = 0.91), and Fun (α = 0.90). The 15-GAIN is a five-factor construct that measures players’ preferences in gameplay activity types of Aggression (α = 0.88), Caretaking (α = 0.83), Coordinate (α = 0.83), Exploration (α = 0.85), and Management (α = 0.88). The GAIN was applied in the study as it informs us about game choice and game enjoyment [16], whereas the IMG instrument has been developed and validated precisely for investigating approach motivations to play video games at large. We made an exploratory cluster analysis with Stata 16.2 software in order to explore how the motivational factor of Fun was related to other play motives (i.e., approach factors) and furthermore to gameplay preferences (i.e., enjoyment factors), genre play habits, and player demographics. For this purpose, we generated factor sum variables for each of the five IMG dimensions and proceeded to do a cluster analysis based on these motive factors. The number (k) of clusters was identified by investigating scree plots generated from the within sum of squared (WSS) and its logarithm [log(WSS)] for cluster solutions between 2 and 20 clusters. A solution of eight clusters was found to be the most prominent, and therefore we conducted a nonhierarchical k-means cluster analysis with Euclidean distance for k = 8.

3

Results

As a result of the cluster analysis, it was first found that Fun had the highest mean sum of the five IMG-based motive factors across all clusters. We then standardized the values for all of the five factor sums by each cluster and found that the effect size between the Fun factor and the second most highest ranged from small (d = 0.23) to huge (d = 1.94) and that this approach generated, in principle, player clusters that were very motivated, reasonably motivated, slightly motivated, and amotivated to play videogames. Differently put, by initially generating clusters based on sum variables, the results did not inform us much about the motivational profiles as the k-means procedure identified the clusters based

Discovering the Motivational Constitution of ‘Playing Games for Fun’

43

Table 1. The six-cluster solution. Reporting standardized factor sums and direct factors sum. Immersive Autonomous Challenger Fun-Seeker Social Competitive N

199

623

392

566

229

201

Standardized Relatedness

0.03

−0.84

0.27

0.14

1.43

1.07

Competence

−0.61

0.34

0.75

−0.73

−0.57

0.29

Immersion

1.10

−0.15

0.10

−0.10

0.07

−0.28

Fun

−0.88

−0.12

−0.22

0.85

−0.49

−1.24

Autonomy

0.37

0.76

−0.89

−0.17

−0.43

0.16

Factor sums Relatedness

2.47

1.82

2.14

1.72

3.39

3.29

Competence

2.74

3.39

3.29

1.95

2.49

3.43

Immersion

3.50

2.38

2.16

1.68

2.39

2.43

Fun

3.24

3.66

3.26

3.53

3.29

3.02

Autonomy

3.39

3.53

1.99

2.16

2.50

3.24

Motive average 3.07

2.95

2.57

2.21

2.81

3.08

on how motivated players were instead of being able to generate insight into what motivates them to play games. To overcome this issue, another approach to k-means clustering was taken. Prior to making the analysis, the five factor sum variables were now standardized for each survey participant. By doing so, we excluded from the cluster analysis information about to what degree a participant was motivated to play games in general and based the analysis only on what factors motivated them the most and the least in comparison to their own motivation mean value. The scree plots suggested a solution of six clusters, and a k-means clustering was made accordingly. The six clusters are reported in Table 1. Of the six clusters, only Cluster 4 was identified as based on its players desire to play video games because of Fun. Participants of Cluster 1 were grouped together, because of the Immersion and Autonomy motive whereas Autonomy and Competence were both relatively high motives for the second player cluster. The Competence motive was the most important motivational factor for Cluster 3. And finally both Cluster 5 and Cluster 6 were motivated by Relatedness. Cluster 6 had the lowest value for Fun of the six clusters. However, when we observed the non-standardized factor sums it was revealed that Fun had the highest mean value not only for Cluster 4 but also for Cluster 2. The significance of the Fun motive was evident also when taking into account that Cluster 2 and Cluster 4 were clearly the two largest clusters with 623 and 566 participants. Next we calculated GAIN factor sums, genre playing habit means, game mode preferences, and descriptive statistics for demographics for the six clusters (Table 2). Participants of the Immersive cluster had the highest mean preference for Exploration, Aggression, Caretaking, and Management of the six player clusters.

44

K. Tuuri and J. Vahlo

Table 2. Descriptive GAIN, genre play, game mode, and playtime statistics for the six clusters. Immersive Autonomous Challenger Fun-Seeker Social Competitive N

199

623

392

566

229

201

Aggression

2.71

2.26

2.10

1.81

2.47

2.5

Caretaking

2.69

2.36

2.40

2.11

2.42

2.44

Coordinate

2.49

2.31

2.24

1.87

2.39

2.53

Exploration

3.28

3.02

2.69

2.38

2.87

2.86

Management

2.93

2.64

2.30

2.02

2.56

2.70

GAIN average

2.82

2.52

2.34

2.04

2.54

2.62

Action

2.22

1.70

1.70

1.49

2.02

2.19

Action-adventure

2.44

1.77

1.76

1.50

1.81

2.06

Racing

2.36

1.94

1.98

1.89

2.51

2.46

Puzzle

2.72

3.04

3.08

3.04

2.81

2.78

RPG

2.62

1.88

1.92

1.60

2.28

2.31

Simulation

2.38

1.83

1.77

1.68

1.97

2.11

Sports

2.18

1.65

1.79

1.59

2.02

2.31

Strategy

2.45

1.87

1.91

1.69

2.38

2.50

Platformer

2.33

1.78

1.89

1.71

1.95

2.09

Genre average

2.21

1.72

1.77

1.58

1.94

2.09

Single-player PC/console games 2.98

3.15

2.68

2.70

2.66

2.78

Single-player mobile games

3.05

3.32

3.04

3.24

2.97

2.94

Co-op PC/console games

2.76

2.26

2.24

2.13

2.67

2.77

Multiplayer PC/console games

2.54

2.15

2.03

1.88

2.64

2.71

Multiplayer mobile games

2.40

2.09

2.08

1.88

2.52

2.53

PC play hours/week

2.91

2.62

2.17

2.11

3.69

3.18

Console play hours/week

1.79

1.36

0.82

0.90

0.72

1.19

Mobile play hours/week

3.53

3.79

4.36

4.23

4.26

3.55

They also had the highest mean value for genre play, meaning that they played more genres than players of the other clusters. They played action games, actionadventure games, role-playing games, simulations, and platform games more than other players. The Immersive players had the highest weekly mean play hours on console play (1.8 h), and lowest mean age (35.9 years, 57% male). The cluster was under-represented in the Finnish sample. The Autonomous players had the second highest preference average for Exploration, and they reported the highest preference for single-player computer and console play, and single-player mobile play. The puzzle genre was their favorite. They were the oldest player cluster (42.6 years, 50% male), and they were overrepresented in Finland. The Challenger player type (40.1 years, 54% female) had the highest preference for the puzzle genre and they played mobile games each week for 4.4 h, which was the highest value of the player clusters. They were equally represented in both countries. The Fun-Seeker was only motivated to play because of Fun. They also did not prefer any of the gameplay activity types but instead had the lowest preferences

Discovering the Motivational Constitution of ‘Playing Games for Fun’

45

for all of the five factors of the six player clusters. Furthermore, they had the lowest game genre average, but they played puzzle games almost as much as the Challenger type. They also had the second highest weekly play hours for mobile gaming, and the second highest preference average for single-player mobile games after the Autonomous type. A total of 57% of this player type were female players. It was equally represented in both countries, and the mean age of the cluster was 40.5 years. Finally, the Social and the Competitive player clusters were both motivated by Relatedness. They both enjoyed Exploration to a similar extent, and they both reported high scores for the racing game genre. The Competitive also had a relatively high score for sports games, and the highest score for strategy games of the six clusters. They both enjoyed multiplayer computer and console games more than the other clusters. The Social player type had the highest weekly computer play hours of the clusters, followed by the second highest hours of the Competitive type. Both of these clusters were over-represented in South-Korea. A total of 56% of the Social cluster were male players, but this was overshadowed by the Competitive player type which consisted of 60% of males. If we consider together the two clusters (Autonomous and Fun-Seeker ) that had the highest mean sums for the Fun motive, we can note that these two clusters covered 53.8% of all players included in the study. These two clusters shared similar game preferences and interests as both of them preferred the puzzle genre and single-player experiences. On the same note, the clusters that had the highest Relatedness motive for game play had notably lower Fun motive scores. This raises a question about a possible tension between the social motive to play when compared to playing just for Fun.

4

Concluding Statements

To a large part, the results of the present investigation seem to agree with the general expectations that playing games is fun, people play games because of the fun, and that such fun associates with the voluntariness of gameplay. In the results of the first cluster analysis, fun in particular appeared as an allencompassing and ubiquitous motivational orientation, which would arguably incorporate expectations of gameplay gratifications and motivational factors in a diffuse manner. Due to this ubiquitous nature, following Crawford’s [2] line of thought, it seems difficult to treat fun as the same level of motive as the more discrete need-based IMG motives overtaken by it. Rather, fun could be considered as more like a general game gratification orientation. The second cluster analysis, however, offered much more defined results revealing a six cluster solution, in which only one of the clusters (25.6% of the players) was solely based on a general fun-seeking game gratification orientation. Thus, the players represented by this Fun-seeker cluster appear purely as gratification seekers, while with the other clusters, fun appears to be more defined and needs-based. The results support Tamborini et al.’s [13] model that makes a distinction between the deeper, SDT needs-based and the more superficial and

46

K. Tuuri and J. Vahlo

hedonistic-driven orientations towards enjoyment. Furthermore, the results are in line with Ryan et al.’s [8] conceptual distinction between hedonistic and eudaimonic approaches to human well-being, which in particular proposes that the latter approach indeed can be characterized in SDT-based motivational terms. The question remains: what are the qualities of game gratification that are preferred by the Fun-Seekers? Our results indicate that they prefer logical problem solving and other game mechanics and interaction types typical of mobile puzzle games. In future studies this issue could be further investigated by including scrutiny of gameplay challenge dimensions, as well as focusing more on the performative ways of pleasure constitution. The latter approach could, for example, include a particular focus on the vitality forms [11] of the actions of gameplay as an immediate arousal of pleasure within sensory-motor interaction experience.

References 1. Bisson, C., Luckner, J.: Fun in learning: The pedagogical role of fun in adventure education. J. Exp. Educ. 19(2), 108–112 (1996) 2. Crawford, C.: The art of computer game design. Osborne/McGraw-Hill, Berkeley, California (1984) 3. Csikszentmihalyi, M.: Flow: the psychology of optimal experience. Harper & Row, New York (1990) 4. Goffman, E.: Encounters. Two Studies in the Sociology of Interaction. Mansfield Centre, CT: Martino Publishing (1961) 5. Huizinga, J.: Homo Ludens: A Study of the Play-Element in Culture. Routledge, United Kingdom (1949) 6. Prensky, M.: Digital Game Based Learning. McGraw-Hill, New York (2001) 7. Ryan, R., Deci, L.: Intrinsic and extrinsic motivations: classic definitions and new directions. Contemp. Educ. Psychol. 25, 54–67 (2000) 8. Ryan, R., Huta, V., Deci, L.: Living well: A self-determination theory perspective on eudaimonia. J. Happiness Stud. 9(1), 139–170 (2008) 9. Sherry, J.L., Lucas, K., Greenberg, B.S., Lachlan, K.: Video game uses and gratifications as predictors of use and game preference. In: Vorderer, P., Bryant, J. (eds.) Playing Video Games. Motives, Responses, and Consequences, pp. 213–224. New York: Routledge, Taylor & Francis Group (2006) 10. Sicart, M.: Play Matters. MIT Press, Cambridge, MA, London (2014) 11. Stern, D.N.: Forms of vitality: Exploring dynamic experience in psychology, the arts, psychotherapy, and development. Oxford University Press (2010) 12. Suits, B.: The Grasshopper. Games, Life and Utopia. Peterborough: Broadview Press (1978) 13. Tamborini, R., Bowman, N.D., Eden, A., Grizzard, M., Organ, A.: Defining media enjoyment as the satisfaction of intrinsic needs. J. Commun. 60(4), 758–777 (2010) 14. Tisza, G., Markopoulos, P.: FunQ: measuring the fun experience of a learning activity with adolescents. Curr. Psychol. 1–21 (2021). https://doi.org/10.1007/ s12144-021-01484-2 15. Vahlo, J., Hamari, J.: Five-Factor Inventory of Intrinsic Motivations to Gameplay (IMG). In Proceedings of the 52nd Hawaii International Conference on System Sciences, pp. 2476–2485 (2019) 16. Vahlo, J., Smed, J., Koponen, A.: Validating gameplay activity inventory (GAIN) for modeling player profiles. User Model. User Adap. Inter. 28, 425–453 (2018)

Towards an Understanding of How Players Make Meaning from Post-Play Process Visualizations Erica Kleinman1(B) , Jennifer Villareale2 , Murtuza Shergadwala1 , Zhaoqing Teng1 , Andy Bryant1 , Jichen Zhu3 , and Magy Seif El-Nasr1 1

University of California Santa Cruz, Santa Cruz, CA, USA {emkleinm,mshergad,zhteng,mseifeln}@ucsc.edu, [email protected] 2 Drexel University, Philadelphia, PA, USA [email protected] 3 IT University of Copenhagen, Copenhagen, Denmark [email protected]

Abstract. Player-facing, retrospective gameplay visualizations help players track progress and learn from others. Visualizations of a user’s step by step process, may be able to advance retrospective visualization. However, we currently do not know how players make meaning from process visualizations of game data. In this work, we take a first step towards addressing this gap by examining how players make meaning from process visualizations of other players’ gameplay. We identify two interpretation methods comprised of six techniques and discuss what these results mean for future use of player-facing process visualizations. Keywords: Post-game visualization · Process visualization · Visualization interpretation · Interpretation techniques · User studies

1

Introduction

The last decade has seen an increased interest in player-facing post-game visualizations [5,5,15,21,29,43,44,50]. Recent work has, however, revealed that existing visualizations do not often provide enough causal information for players to make connections between their actions and the outcomes they experienced [23]. Process visualizations, which present human process as a sequence of actions taken [3,30,37,41,42,45,46], appear well suited to preserving causal information. Player-facing, post-play process visualizations are, however, rare and typically included as a secondary feature to a visualization of another type [1,24,32]. As such, they are rarely the focus of research, and we know little about how players extract meaning from them, which is necessary to ensure that we design and implement them in appropriate and effective ways. c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 47–58, 2022. https://doi.org/10.1007/978-3-031-20212-4_4

48

E. Kleinman et al.

In this paper, we take the first steps to address this gap by examining how players make sense of post-play process visualizations of others’ gameplay in the context of an educational game. In particular, our research question is: “What interpretation techniques do players use to make sense of process visualizations of others’ gameplay?” We chose to focus on having participants interpret other players’ data due to the significant role that reviewing the gameplay of others’ plays in learning how to play games [36]. To answer this question, we conducted a qualitative study with 13 players of the game Parallel [49], prompting them to make sense of other players’ gameplay through a process visualization. Results revealed six interpretation techniques that players leverage to extract meaning from game data. We also identified two general sense-making methods for postplay process visualizations: the induction method and the framing method. Based on these results, we present and discuss four general design implications that should be considered in future design and development.

2

Related Work

To date, most player-facing gameplay visualizations feature either aggregate [5,15] or spatio-temporal data [1,24,44]. Aggregate visualization techniques use visual elements such as percentages, graphs, and charts [5,15,29]. However, such visualizations do not preserve granular strategic information [20], making it difficult for players to determine where they may have made a mistake. Spatiotemporal visualizations, in contrast, present granular, action by action, data, superimposed atop a game map [1,22,24,44]. However, when scaled, spatiotemporal visualizations often remove granular data, and instead focus on movement over time [44]. While informative, these visualizations loses much of the causal information. These drawbacks created a space that process visualizations began to fill. In other domains, review of process has been valuable for optimizing human workflows [40,41] and learning [27,28]. In games, graph-based process visualizations are used for game analytics and user experience research [2,3,9,16– 18,20,30,45]. Player-facing process visualizations, in contrast, are typically designed as timelines depicting the ordering of actions taken over the course of a game [1,22,24,32]. However, these timelines are often secondary features attached to another visualization system [1,24]. As such they are often not the focus of research. This results in a lack of knowledge about how players extract insights from process visualizations, which is necessary for informed design. Understanding how users make sense of data is pivotal for the design of player-facing visualizations. While there is a lot of work in InfoVis [10,34,35,47], in the context of games such work is sparse. Previous work investigated this question in spatio-temporal visualization [22] and demonstrated that making meaning from visualized game data requires unique investigation. However, to the best of our knowledge, no one has specifically investigated this question in the context of process visualizations.

Towards an Understanding of Process

3

49

Methodology

Parallel is a puzzle game designed to teach parallel programming concepts [19, 39,49]. We chose to conduct our research with an educational game, because retrospective visualizations are already common in digital learning contexts [4, 43,50]. We chose Parallel for this study as it is complex enough for players to demonstrate various approaches to solving problems, yet simple enough for players to become comfortable with gameplay quickly. Using the visualization tool Glyph [30], we generated a process visualization (see Fig. 1) based on 15 key strategic Parallel gameplay actions. Each node in Glyph’s network graph represented a different in-game action and a link between two nodes indicated that at least one player in the community transitioned between those two actions. Individual player trajectories within this visualization can be highlighted as seen in Fig. 1.

Fig. 1. An screen shot of the process visualization, with player 8’s sequence highlighted.

13 undergraduate computer science students were recruited from universities in the United States, as they represent the target population for Parallel [49]. Gameplay took from 30 to 60 min. Players then signed up for data-driven retrospective interviews [8] conducted over Zoom. 3.1

Interview Protocol

Prompt Design. Based on previous work [22,34,38] we recognized two types of interpretation techniques: interaction techniques used to extract information

50

E. Kleinman et al.

from the visualization and cognitive techniques used to make sense out of that information. To ensure that we elicited techniques in both categories, we developed two prompts: 1. Could you describe this player’s actions using the visualization? (“interaction prompt”) 2. Can you say why you think the player played the way they did? (“cognitive prompt”) Procedure. A slide deck was prepared and presented to each participant during the interview. One researcher led the interview, screen-sharing the slide deck, while two others remained silent and recorded, in text, what the participant said. The first slide contained a visualization of the participant’s own data. On this slide, the lead researcher gave the participant basic instructions on how to read the visualization. The next slide contained a visualization of another participant, who played similarly to the interviewee. The last slide contained a visualization of another participant, who played differently than the interviewee. While displaying the second and third slides, the lead researcher asked the prompts described above. Interviews lasted about 30 min and participants received a 50$ gift card. 3.2

Data Analysis

Interview data was analyzed using a two-step, iterative thematic analysis protocol [13,33]. The first step of the analysis identified the specific interpretation techniques that players used. Two researchers, separately, performed open coding on the interview responses. The unit of analysis was a player’s response to a prompt. They then met and discussed their initial codes to generate a code book of six interpretation techniques. The researchers then performed an interrater reliability check using Cohen’s Kappa [7] on 30% [6] of the data. The codes achieved an IRR score of .87, indicating very strong agreement [25]. The second step identified the overall process of making sense of the data. The two researchers separately analyzed each prompt response and marked which of the six interpretation strategies were used and in what order. The researchers then reconvened and discussed their findings. They identified two methods for engaging the interpretation techniques and performed a second inter-rater reliability check, again on 30% of the data set [6]. The method codes achieved an IRR score of .74, indicated strong agreement [25]. One researcher then labeled the remainder of the data set with the method codes.

4

Results

The six interpretation techniques are shown in Table 1.

Towards an Understanding of Process

51

Table 1. The six interpretation techniques identified based on analysis of players’ interaction with the community visualizations and brief definitions. Interpretation Technique

Definition

Reading the Visualization to Collect Information

The participant read the data that appeared in the visualization but did not extrapolate on it

Identifying Patterns to Inform Inferences: Sequential Pattern

The participant identified a pattern related to the ordering of data-points

Identifying Patterns to Inform Inferences: Frequency Pattern

The participant identified a pattern related to the amount of data-points

Making a Comparison to Guide Pattern Identification

The participant compares the data of the other player to their own experience to better make sense and extract patterns

Making an Inference to Understand the Other Player: Approach or Strategy

The participant makes an inference regarding the subject’s intentions behind the actions they took

Making an Inference to Understand the Other Player: Understanding or Expertise

The participant makes an inference regarding the subject’s knowledge or comprehension of the gameplay

Reading the Visualization to Collect Information. Participants would try to collect information from the visualized data as a precursor to making connections between data points. For example, P9 reads the trajectory of another player, stating “They do start, test passed, sub failed, they place the semaphore, then maybe they toggle it, they place the signal, they link, maybe they move it around.” Notably, reading the visualization would often encompass a readthrough of the entire sequence, suggesting that participants were engaging this method to gain a holistic overview of the data. This is illustrated by P0 who said “it looks like they placed and moved semaphores and placed signals, linked some together, ran a submission, stopped the submission, placed another semaphore, maybe moved it again, toggled it, and then maybe toggled a different one, ran it, a test passed, and then the submission passed.” Notably, participants did not always read the visualization, many jumped to identifying patterns. Identifying Patterns to Inform Inferences. Participants would make general statements about the characteristics of the data. We saw two types of patterns that participants would identify: – Sequential Pattern: Refers to the participant identifying patterns in the ordering of actions. For example, P3 noticed that “[the other player] repeats that process of toggling then placing then linking”. Recognition of these patterns is facilitated by the sequential nature of the visualization and it would likely be harder to recognize such patterns otherwise. – Frequency Pattern: Refers to the participant identifying a pattern regarding the number of actions taken. For example, P0 said: “They move a limited

52

E. Kleinman et al.

number of times, but they ran the submission a lot because it looks like they stopped it a lot.” We observed that pattern recognition would lead to inferences regarding the players who generated the data. For example, P8 described “They seem to jump back and forth a lot. They were probably thinking through a lot of their placement and movement.” Making a Comparison to Guide Pattern Identification. Comparison did not always occur, but when it did, players would typically compare patterns in peers’ gameplay to patterns in their own. For example, P0 said “They use the stop submission button, that’s interesting, I don’t think I used it at all.” Often, participants would use comparison as a way to guide the identification of additional patterns. This is well illustrated by P1 “Once they laid down a solution they would test it and see if it failed or not. Whereas I don’t remember doing as much testing.” Here, the participant has identified a pattern in which the subject would lay down a solution then test it. They compare this to their own gameplay, in which they did not test as much. Such comparison can help them identify more patterns (what else did the other player do differently?) and begin to generate a more formal inference. Making an Inference to Understand the Other Player. As discussed above, inferences were informed by identified patterns within the data, which were sometimes guided by comparison. We observed inferences to be focused primarily on the player, which differs from previous work [22]. We observed two types of inferences: – Approach or Strategy: Refers to the participant making an inference about subject’s plan execution. For example P10 said “They saw that the test passed so their aim was to try and generalize the solution.” Here, P10 infers a strategic decision that the player made (trying to generalize their solution) as a way of explaining an observed pattern in the data (that the player did not immediately submit after their test passed and instead took other actions). – Understanding or Expertise: Refers to the participant making an inference about what the subject knows about the task or subject. For example P11 said “I would say they probably came in with a good idea about how they were going to do the level before they started playing [since] they’re very calculated, they rarely jump back and forth between states.” Here, the participant has developed an image in their mind regarding the expertise of the other player (that they had a good idea of what they were going to do) that can be used to explain an observed pattern in their data (that they rarely go back to previously visited states). 4.1

Sense-Making Methods

The interpretation techniques connect to one another to form a process for making sense of the data. We refer to this process as a sense-making method. We

Towards an Understanding of Process

53

identify two general methods for sense-making for post-play process visualizations for games, seen in Fig. 2 and described in detail below:

Fig. 2. The sense-making methods we observed in terms of the ordering of interpretation techniques.

Induction Method: This method represents an approach in which players began their sense-making process by reading the visualization. They would then identify patterns in the data, and use comparison, if necessary, to generate an understanding of gameplay events. This would culminate in an inference about the other player. An example of this method is demonstrated by P7: first, they read the visualization, stating “Ran a test and it passed then worked to place the items in one sequence, and then the test failed, and then in another they stopped it again.” They follow this with recognition of a frequency pattern, stating “It looks like they placed a lot”. Finally, they offer an explanation, stating “they probably deleted [the signals and semaphores] instead of moving them.” Framing Method: When participants used this method, they first made inferences about the other player, often based on visually apparent details, e.g., length of the trajectory. They would then switch to collecting information, first reading the visualization, then using one or both pattern identification techniques and comparison, to generate hypotheses that justified and supported their initial inference. An example of this method is demonstrated by P12: they begin with an inference of the other player’s strategy (or lack thereof), stating “I would think that this player kind of did stuff at random, I’m not sure if there was a process that they used.” They follow this by reading the visualization to collect information, stating “It seems like [they’re] going from start and then placing a semaphore [then] going from test passed to stopping submission and moving a signal”. They follow this with identification of a sequential pattern (or lack

54

E. Kleinman et al.

thereof), stating “It doesn’t look like this graph had a lot of iterative processes. It’s a little jumbled up.”

5

Discussion and Implications

It becomes apparent that inferences facilitate players’ ability to extract actionable insights from data. This finding is similar to what has been discussed in InfoViz work regarding mental models of data [26,47,48]. Unlike InfoViz work, the inferences here inform a mental model of the individual who produced the data rather than the data itself, similar to what was seen by Kleinman et al. in their study of spatio-temporal post-play visualization [22]. Further, in this work we see the presence of a sense-making method that begins with the inference and then collects data to enforce it. This may have been encouraged by the nature of the visualization, from which surface level information, such as length of trajectory, could be quickly extracted and used to reach a preemptive conclusion. This suggests that process visualizations, which present data in a holistic manner, may encourage players to make assumptions about the data up front. However, there is the very real possibility that these up-front assumptions can lead to inaccurate inferences. Thus, process visualizations for postplay analysis should consider incorporating design elements that can inform players’ up-front assumptions and guide them towards correct initial inferences. One way to accomplish this could be the grouping or labeling actions inside a visualization to indicate what they mean. Further, while previous work discusses users adjusting frames and hypotheses [22,26], participants in our study who used the framing method did not make adjustments. In fact, it seemed that they rarely uncovered information that they recognized as contradictory to their inference. Based on our results, we hypothesize two reasons for this. The first reason is related to the participants’ familiarity with the game. In our study, participants had no prior experience with Parallel. As a result, they likely lack the domain knowledge necessary to recognize gameplay strategies in the data. Thus, process visualizations may aid players best if they are not displayed until the player has become more familiar with the game. The second reason is related to the abstraction of the data. The presentation of the gameplay data as a trajectory of actions may have been too abstract. Including game state, recognized as important to understanding context [22], information in the process visualization could have helped players better understand what they were observing. Thus, retrospective process visualizations should consider incorporating game state information, to ensure that players are able to correctly interpret the context behind each action. This implication, along with the previous one, can help ensure that the player is equipped to correct misunderstandings about the data. Additionally, the inclusion of comparison, as shown in the results, is not discussed in the previous work by Kleinman et al. [22], where players were not

Towards an Understanding of Process

55

shown their own data. This suggests that the inclusion of a player’s own data is likely to spark comparison between themselves and others. Using comparison between self and others has been explored in the domain of personal informatics, though usually within the context of a user understanding their own data through the comparison [11,31]. Here the comparison was used to understand the other player, as finding the differences in how the other player behaved compared to oneself gave participants an anchor point to begin understanding the rest of their experience. This suggests that process visualizations can leverage comparison to help players more quickly identify connected patterns and reach inferences. Thus, process visualizations in post-play contexts should consider highlighting how the player’s own data compares to and differs from the data of the subject of analysis. This does, however, raise questions about the potential risks of prompting comparison among players, as previous work has demonstrated that players who under-perform can become discouraged when prompted to compare themselves to high-performing players [12]. Thus, process visualizations may wish to only permit comparison against other players with similar skill levels or quality of performance.

6

Limitations

We acknowledge that this study was performed on a small sample size (n = 13). However, we did see saturation in the data at 7 participants and argue that this sample size aligns with those seen in similar work [14,22]. We additionally recognize that we only looked at a single game, only at the analysis of others’ data, and that the nature of the visualization itself likely influenced our results. As such, more work is needed to ensure the generalizability of the findings. With this in mind we present this work as a first step towards understanding how players make sense of process visualizations of others’ data during post-play analysis.

7

Conclusion

In this work, we take a first step towards understanding how players make sense of process visualizations during post-play analysis. Through a 13-participant qualitative user study, we identified six interpretation techniques that players used to make sense of process visualizations and two methods for sense-making. We discuss the implications of these findings on the use of player-facing process visualizations in post-play analysis. Acknowledgements. This work is supported by the national Science Foundation (NSF) under Grant #1917855. The authors would like to thank all past and current members of the project.

56

E. Kleinman et al.

References 1. Afonso, A.P., Carmo, M.B., Gon¸calves, T., Vieira, P.: Visualeague: player performance analysis using spatial-temporal data. Multimedia Tools Appl. 78(23), 33069–33090 (2019) 2. Ahmad, S., et al.: Modeling individual and team behavior through spatio-temporal analysis. In: Proceedings of the Annual Symposium on Computer-Human Interaction in Play, pp. 601–612 (2019) 3. Andersen, E., Liu, Y.E., Apter, E., Boucher-Genesse, F., Popovi´c, Z.: Gameplay analysis through state projection. In: Proceedings of the Fifth International Conference on the Foundations of Digital Games, pp. 1–8 (2010) 4. Bodily, R., et al.: Open learner models and learning analytics dashboards: a systematic review. In: Proceedings of the 8th International Conference on Learning Analytics and Knowledge, pp. 41–50 (2018) 5. Bowman, B., Elmqvist, N., Jankun-Kelly, T.: Toward visualization for games: theory, design space, and patterns. IEEE Trans. Visual Comput. Graphics 18(11), 1956–1968 (2012) 6. Campbell, J.L., Quincy, C., Osserman, J., Pedersen, O.K.: Coding in-depth semistructured interviews: problems of unitization and intercoder reliability and agreement. Sociol. Methods Res. 42(3), 294–320 (2013) 7. Cohen, J.: A coefficient of agreement for nominal scales. Educ. Psychol. Measur. 20(1), 37–46 (1960) 8. El-Nasr, M.S., Durga, S., Shiyko, M., Sceppa, C.: Data-driven retrospective interviewing (ddri): a proposed methodology for formative evaluation of pervasive games. Entertainment Comput. 11, 1–19 (2015) 9. El-Nasr, M.S., Nguyen, T.H.D., Canossa, A., Drachen, A.: Game Data Science. Oxford University Press (2021) 10. Endert, A., Hossain, M.S., Ramakrishnan, N., North, C., Fiaux, P., Andrews, C.: The human is the loop: new directions for visual analytics. Journal of intelligent information systems 43(3), 411–435 (2014) 11. Epstein, D., Cordeiro, F., Bales, E., Fogarty, J., Munson, S.: Taming data complexity in lifelogs: exploring visual cuts of personal informatics data. In: Proceedings of the 2014 Conference on Designing Interactive Systems, pp. 667–676 (2014) 12. Esteves, J., Valogianni, K., Greenhill, A.: Online social games: the effect of social comparison elements on continuance behaviour. Inform. Manage. 58(4), 103452 (2021) 13. Gavin, H.: Thematic analysis. Understanding research methods and statistics in psychology, pp. 273–282 (2008) 14. Halabi, N., Wallner, G., Mirza-Babaei, P.: Assessing the impact of visual design on the interpretation of aggregated playtesting data visualization. In: Proceedings of the Annual Symposium on Computer-Human Interaction in Play, pp. 639–650 (2019) 15. Hazzard, E.: Data visualization in games (2014). https://erikhazzard.com/blog/ game-development/data-visualization-in-games 16. Horn, B., Hoover, A.K., Barnes, J., Folajimi, Y., Smith, G., Harteveld, C.: Opening the black box of play: Strategy analysis of an educational game. In: Proceedings of the 2016 Annual Symposium on Computer-Human Interaction in Play, pp. 142– 153. ACM (2016)

Towards an Understanding of Process

57

17. Javvaji, N., Harteveld, C., Seif El-Nasr, M.: Understanding player patterns by combining knowledge-based data abstraction with interactive visualization. In: Proceedings of the Annual Symposium on Computer-Human Interaction in Play (2020) 18. Jemmali, C., Kleinman, E., Bunian, S., Almeda, M.V., Rowe, E., Seif El-Nasr, M.: Maads: Mixed-methods approach for the analysis of debugging sequences of beginner programmers. In: Proceedings of the 51st ACM Technical Symposium on Computer Science Education, pp. 86–92 (2020) 19. Kantharaju, P., Alderfer, K., Zhu, J., Char, B., Smith, B., Ontan´ on, S.: Tracing player knowledge in a parallel programming educational game. arXiv preprint arXiv:1908.05632 (2019) 20. Kleinman, E., et al.: “and then they died”: Using action sequences for data driven, context aware gameplay analysis. In: International Conference on the Foundations of Digital Games. FDG ’20, Association for Computing Machinery, New York, NY, USA (2020) 21. Kleinman, E., El-Nasr, M.S.: Using data to “git gud”: A push for a player-centric approach tothe use of data in esports (2021) 22. Kleinman, E., Preetham, N., Teng, Z., Bryant, A., Seif El-Nasr, M.: “what happened here!?” a taxonomy for user interaction with spatio-temporal game data visualization. Proceedings of the ACM on Human-Computer Interaction 5(CHI PLAY), 1–27 (2021) 23. Kleinman, E., Shergadwala, M.N., Seif El-Nasr, M.: Kills, deaths, and (computational) assists: Identifying opportunities for computational support in esport learning. In: CHI Conference on Human Factors in Computing Systems. CHI ‘22, Association for Computing Machinery, New York, NY, USA (2022). https://doi. org/10.1145/3491102.3517654 24. Kuan, Y.T., Wang, Y.S., Chuang, J.H.: Visualizing real-time strategy games: The example of starcraft ii. In: 2017 IEEE Conference on Visual Analytics Science and Technology (VAST), pp. 71–80. IEEE (2017) 25. Landis, J.R., Koch, G.G.: The measurement of observer agreement for categorical data. biometrics, pp. 159–174 (1977) 26. Lee, S., Kim, S.H., Hung, Y.H., Lam, H., Kang, Y.a., Yi, J.S.: How do people make sense of unfamiliar visualizations?: A grounded model of novice’s information visualization sensemaking. IEEE Trans. Visualization Computer Graphics 22(1), 499–508 (2015) 27. Lin, X., Hmelo, C., Kinzer, C.K., Secules, T.J.: Designing technology to support reflection. Educa. Tech. Research Dev. 47(3), 43–62 (1999) 28. Loh, B., Radinsky, J., Russell, E., Gomez, L.M., Reiser, B.J., Edelson, D.C.: The progress portfolio: Designing reflective tools for a classroom context. In: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, pp. 627–634 (1998) 29. Medler, B.: Player dossiers: Analyzing gameplay data as a reward. Game Studies 11(1) (2011) 30. Nguyen, T.H.D., El-Nasr, M.S., Canossa, A.: Glyph: Visualization tool for understanding problem solving strategies in puzzle games. In: FDG (2015) 31. Puussaar, A., Clear, A.K., Wright, P.: Enhancing personal informatics through social sensemaking. In: Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems, pp. 6936–6942 (2017) 32. RPGLogs, L.: Warcraft logs (2013–2021). https://www.warcraftlogs.com/ 33. Salda˜ na, J.: The coding manual for qualitative researchers. Sage (2015)

58

E. Kleinman et al.

34. Sandouka, K.: Interactive visualizations: A literature review. MWAIS 2019 Proceedings. p. 8 (2019) 35. Shneiderman, B.: The eyes have it: A task by data type taxonomy for information visualizations. In: Proceedings 1996 IEEE Symposium on Visual Languages, pp. 336–343. IEEE (1996) 36. Sj¨ oblom, M., Hamari, J.: Why do people watch others play video games? an empirical study on the motivations of twitch users. Comput. Hum. Behav. 75, 985–996 (2017) 37. Temkin, B.D.: Mapping the customer journey. Forrester Res. 3, 20 (2010) 38. Valiati, E.R., Pimenta, M.S., Freitas, C.M.: A taxonomy of tasks for guiding the evaluation of multidimensional visualizations. In: Proceedings of the 2006 AVI Workshop on Beyond time and Errors: Novel Evaluation Methods for Information Visualization, pp. 1–6 (2006) 39. Valls-Vargas, J., Zhu, J., Onta˜ n´ on, S.: Graph grammar-based controllable generation of puzzles for a learning game about parallel programming. In: Proceedings of the 12th International Conference on the Foundations of Digital Games, pp. 1–10 (2017) 40. Van Der Aalst, W.: Process mining: overview and opportunities. ACM Trans. Manage. Inform. Syst. (TMIS) 3(2), 1–17 (2012) 41. Van Der Aalst, W., et al.: Process Mining Manifesto. In: Daniel, F., Barkaoui, K., Dustdar, S. (eds.) BPM 2011. LNBIP, vol. 99, pp. 169–194. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-28108-2 19 42. van der Aalst, W., Adriansyah, A., van Dongen, B.: Causal nets: a modeling language tailored towards process discovery. In: Katoen, J.-P., K¨ onig, B. (eds.) CONCUR 2011. LNCS, vol. 6901, pp. 28–42. Springer, Heidelberg (2011). https://doi. org/10.1007/978-3-642-23217-6 3 43. Villareale, J., F. Biemer, C., Seif El-Nasr, M., Zhu, J.: Reflection in game-based learning: A survey of programming games. In: International Conference on the Foundations of Digital Games, pp. 1–9 (2020) 44. Wallner, G., Kriglstein, S.: Visualizations for retrospective analysis of battles in team-based combat games: A user study. In: Proceedings of the 2016 Annual Symposium on Computer-Human Interaction in Play, pp. 22–32 (2016) 45. Wallner, G.: Play-graph: A methodology and visualization approach for the analysis of gameplay data. In: FDG, pp. 253–260 (2013) 46. White, S.A.: Introduction to bpmn. Ibm Cooperation 2(0), 0 (2004) 47. Yi, J.S., ah Kang, Y., Stasko, J., Jacko, J.A.: Toward a deeper understanding of the role of interaction in information visualization. IEEE transactions on visualization and computer graphics 13(6), 1224–1231 (2007) 48. Yi, J.S., Kang, Y.a., Stasko, J.T., Jacko, J.A.: Understanding and characterizing insights: how do people gain insights using information visualization? In: Proceedings of the 2008 Workshop on BEyond time and errors: novel evaLuation methods for Information Visualization, pp. 1–6 (2008) 49. Zhu, J., et al.: Programming in game space: how to represent parallel programming concepts in an educational game. In: Proceedings of the 14th International Conference on the Foundations of Digital Games, pp. 1–10 (2019) 50. Zhu, J., El-Nasr, M.S.: Open player modeling: Empowering players through data transparency. arXiv preprint arXiv:2110.05810 (2021)

Entertainment Tools and Methods

Plot Composition by Mapping Situation Calculus Schemas into Petri Net Representation Edirlei Soares de Lima1,2(B) , Antonio L. Furtado3 and Marco A. Casanova3

, Bruno Feijó3

,

1 IADE, Universidade Europeia, Av. D. Carlos I 4, 1200-649 Lisbon, Portugal

[email protected]

2 UNIDCOM/IADE, Av. D. Carlos I 4, 1200-649 Lisbon, Portugal 3 Department of Informatics, PUC-Rio, R. Marquês de São Vicente 225, Rio de Janeiro, Brazil

{furtado,casanova}@inf.puc-rio.br

Abstract. In this paper we propose a new plot composition method based on situation calculus and Petri net models, which are applied, in a complementary fashion, to a narrative open to user co-authorship. The method starts with the specification of situation calculus schemas, which allow a planning algorithm to check if the specification covers the desired cases. A Petri net is then automatically derived from the schemas in a second phase, guiding interactive plot generation and dramatization. The applicability of the proposed method is validated through the implementation of an interactive storytelling system capable of representing the generated Petri net models using 2D graphics and animations. Keywords: Petri net · Situation calculus · Interactive storytelling · Plot generation · Dramatization

1 Introduction Readers enjoy a far more pleasant experience with narratives in which they are invited to participate as co-authors. This claim is convincingly expressed by Umberto Eco [8], when talking of “open works” and “works in movement,” i.e., works that deliberately leave decisions on the meaning of specific passages to the care of the reader. However, this ideal, which is hard to satisfy in book format, only now is truly reachable through interactive narratives developed for digital environments. The most popular approach to interactive narrative, especially in narrative-driven games (e.g., Heavy Rain (2010) and Detroit Become Human (2018) by Quantic Dream), is the branching technique (also known as branching path stories [16]). In this technique, the player makes a choice at each branching point. The writer usually builds a rigid structure of branching points through a manually authored process without any consistency check. User decisions involve choosing which way to proceed at branching points, thus leading, knowingly or not, to a subsequent outcome. To identify such branching points, designers usually use Petri nets [3, 23], a graphically structured modeling technique for dynamic systems [22]. © IFIP International Federation for Information Processing 2022 Published by Springer Nature Switzerland AG 2022 B. Göbl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 61–75, 2022. https://doi.org/10.1007/978-3-031-20212-4_5

62

E. S. de Lima et al.

A more suitable and robust approach is to treat a narrative as an application process where, instead of a single linear plot, the designers (authors) define a fixed repertoire of predefined event-producing operations containing preconditions and post-conditions. These conditions enforce the intended conventions and preserve consistency when the users (readers performing as co-authors) are allowed to decide. Postconditions consist of facts asserted or retracted as the consequence of executing the operator. And as we argue in this paper, situation calculus [15] is a suitable modeling strategy that can exploit backward-chaining plan generation based on the operators above mentioned (e.g., using STRIPS [11]) to show what specific plots can emerge. This paper presents a new plot composition method that combines both modeling techniques in a complementary fashion. We consider Petri nets from the perspective of event logs, as proposed by Wil van der Aalst [1]. Our method starts with specifying situation calculus schemas for a chosen process application. It then automatically derives a Petri net representation from these schemas, which is informative enough to be employed for interactive plot generation and dramatization. The method is analogously applicable to information system domains, where business transactions can be treated in the same way as narrative plots. The contribution of our work is twofold. First, we shed light on the complementarity between situational calculus and Petri nets. Second, we have taken the process of composing interactive plots to a more robust and semantically consistent level. In our approach, situation calculus is used at specification time, enforcing integrity constraints, and checking if the specification allows all desirable use cases while disallowing undesirable cases. In the final step, the automatically generated Petri net allows visualizing the processes and effectively executing them. The paper is organized as follows. Section 2 discusses related work. Section 3 presents our approach to deriving a Petri net model from a situation calculus model in a narrative domain. Section 4 explores the application of the proposed method in a fully implemented interactive storytelling system. Finally, concluding remarks are the object of Sect. 5.

2 Related Work Situation calculus [15] provides a second-order logic method to formalize state transitions caused by event-producing operations. Petri nets, in turn, are commonly utilized in Process Mining work [1] for obtaining an implicit model of an application, by discovering the partial order requirements prevailing on a significant number of traces extracted from an execution log. For example, in [14], Petri net synthesis is preceded by a preliminary activity mining algorithm. From [1] we borrowed and used in our first experiments [19] the simple introductory case of a request processing application, represented by a Petri net. When specifying the trial by combat application that serves as a running example in the present paper, we were able to end up with a structurally analogous Petri net representation, thus favoring the claim [6] that serious and entertainment applications can be treated by the same modelling formalisms. Research involving situation calculus and Petri net formalisms includes in special [24], which proposes a formal ontology that highlights the correspondence between a

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

63

sequence of actions starting from an initial state in situation calculus, and a sequence of transition firings starting from an initial node in a Petri net. However, the aforementioned work focuses only on the analysis of structural properties of Petri nets and situation calculus models. The relations between Petri nets and automated planning were also explored in previous research, such as [13], where a method to transform planning graphs into Petri nets is presented and used to demonstrate that Petri net unfolding, a form of partial order reduction, can be used to recognize independent planning subproblems. The transformation of Petri net models into planning problems was also explored in [2], which is the inverse of the process discussed in this paper. Plans are modelled as Petri nets in [25] and plans are used to produce workflows in [10] (re-calling that Petri nets can be viewed as a particular form of workflow). Petri nets were also applied in interactive storytelling contexts. Riedl et al. [23] uses a specialized type of Petri net, called colored Petri net, to allow authors to manually model interactive narratives as a process in which multiple players can navigate through different narrative scenes. During dramatization, their system uses an execution algorithm that monitors for situations in which the Petri net fails to account for player actions. When a failure situation is identified, a planning algorithm is used to generate new narrative events to restore the integrity of the Petri net. In this paper, instead of relying on manually authored Petri nets, we focus on the automatic process of mapping situation calculus schemas into Petri net models. Petri nets are also commonly used as modelling tools to design narratives and character behaviors in games. An example of work that employs manually authored Petri nets to represent narrative plots is presented by Balas et al. [3], which uses hierarchical Petri nets to define branching narratives for games. Lee and Cho [17] also proposed a quest generation method for games where quests are modeled as Petri nets, which are activated during a game session according to characters’ goals and the current world state. A similar approach is explored by El-Sattar [9], who uses Petri nets as a state-based model to design narrative plots. The use of Petri nets to model and control individual characters in an interactive storytelling context is also explored by Brom and Abonyi [4], who utilize manually designed Petri nets to represent the narrative of the game. Although Petri nets have been previously explored as a structure to represent narrative plots, most of the previous research focus on the use Petri nets as a modelling tool to allow authors to design interactive narratives. In this paper, we follow a different approach and focus on plot composition by automatically mapping situation calculus schemas into Petri net representations that are suitable for interactive dramatization.

3 From Situation Calculus to Petri Net Models 3.1 The Basic Situation Calculus Model Situation calculus is a logical language used to represent and reason about dynamic worlds, which has been successfully applied to a variety of domains and problems, including narrative generation [6]. According to Kowalski [15], situation calculus, as a logic program, can be compactly expressed by the following two clauses, which define

64

E. S. de Lima et al.

what sentences P hold in the situation result(A, S) that is the result of the transition from state S by an action of type A:

holds(P, result(A, S)) ← happens(A, S) ˄ initiates(A, S, P) holds(P, result(A, S)) ← happens(A, S) ˄ holds(P, S) ˄ ¬terminates(A, S, P) noting that the second clause of this second-order logic formulation avoids the exponential proliferation of first-order logic clauses, which constitutes the so-called frame problem by eliminating the need to specify every class of facts (P) that are not affected by the execution of an operation (A). In turn, these situation calculus clauses suggest an elementary plan-generator, which can be thus expressed in natural language: • A fact F holds if it is true in the initial state; • F holds if it is added as one of the effects of an operation Op, and the preconditions of Op hold at the current state; • F holds after the execution of an operation Op if it did already hold at the current state and if it is not deleted as one of the effects of Op. and then translated into a Prolog program: holds(Fact, [start]) :- initial_state(Fact), !. holds(Fact, [Operation | Current_state]) :- added(Fact, Operation), precond(Operation, Current_state). holds(Fact, [Operation | Current_state]) :- not deleted(Fact, Operation), holds(Fact, Current_state).

Although this elementary program is able to handle overly simple cases, such as the well-known monkey-and-bananas problem, it must be considerably expanded for practical usage, such as proposed in [5, 7]. As a general starting point to apply situation calculus in storytelling domains, one must specify static and dynamic schemas, which include the classes of facts that will eventually populate states, a set of facts describing the initial state, and a fixed repertoire of event-producing operations for performing state changes in conformance with the applicable integrity constraints. Each operation is defined in terms of pre-conditions, which consist of conjunctions of positive and/or negative terms expressing facts, and any number of post-conditions, consisting of facts to be asserted or retracted as the effect of executing the operation (cf. The STRIPS model [11]). As an illustration, we shall concentrate on a simple narrative incident taken and adapted from the film Excalibur, directed, produced, and co-written by John Boorman in 1981. The incident can be thus summarized: Sir Gawain accuses Queen Guinevere of adultery. A trial by combat is announced, there being two candidate knights to claim the Queen’s innocence: Lancelot, a worthy knight, famous for his many victories, and Perceval, who would be no less reputed in the future, but at that time still had little combat experience. The Queen would be vindicated if her defender could defeat the accuser, otherwise she would be condemned. The trial could be reinitiated if a last-minute replacement of defender chanced to occur.

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

65

The entities involved in the incident and their properties are specified through a static schema:1 entity(person, pn). entity(knight, kn). attribute(knight, strength). attribute(knight, loyal). entity(accuser, an). entity(defendant, dn). attribute(defendant, has_defender). entity(defender, dn). entity(challenger, cn). entity(offense, on). relationship(accusation, [defendant, offense]). attribute(accusation, vindicated). attribute(accusation, condemned). relationship(encounter, [challenger, defender]). attribute(encounter, winner).

The event-producing operations are defined in a dynamic schema. For example, the operator for the combat event is defined below, noting that the V parameter identifies the winner, who can be either the accuser or the defending knight, depending on their strength: operation(combat(A, K, D, O, V)). precond(combat(A, K, D, O, V), (accusation(D, O), challenger(A), defender(K), strength(K, Sk), strength(A, Sa), if(Sk > Sa, V = K, V = A))). added(encounter(A, K), combat(A, K, D, O, V)). added(winner([A, K], V), combat(A, K, D, O, V)). added(defendant(D), accuse(A, D, O)).

In order to conduct experiments, an initial state must be introduced, indicating the instances of the entity classes and the specific initial values of their properties: person('Guinevere'). knight('Lancelot'). loyal('Lancelot', true). strength('Lancelot', 200). knight('Perceval'). loyal('Perceval', true). strength('Perceval', 100). knight('Gawain'). loyal('Gawain', false). strength('Gawain', 150). offense(murder). offense(adultery).

Described in this fashion, the well-intentioned but still immature Perceval would stand no chance to defeat Gawain when playing the role of defender. This default result is evidenced in the plan-generated plot below, in which the last parameter of the combat operation (in boldface) indicates the winner:

1 A complete description of the static and dynamic schemas used in this example is available at:

http://www.icad.puc-rio.br/~logtell/petri-net/schemas-trial-by-combat.pdf.

66

E. S. de Lima et al.

accuse(Gawain, Guinevere, adultery), enter_challenger(Gawain, Guinevere, adultery), enter_beginner_defender(Perceval, Guinevere, adultery), combat(Gawain, Perceval, Guinevere, adultery, Gawain), condemn(Guinevere, adultery).

On the contrary, Lancelot had what was required to triumph, thereby establishing the Queen’s innocence: accuse(Gawain, Guinevere, adultery), enter_challenger(Gawain, Guinevere, adultery), enter_worthy_defender(Lancelot, Guinevere, adultery), combat(Gawain, Lancelot, Guinevere, adultery, Lancelot), vindicate(Guinevere, adultery).

Both for serious and for entertainment applications, the situation calculus model leads to the verification, by applying plan-generation, whether the proposed specification allows all desirable use cases and effectively disallows those which transgress the intended conventions [6]. On the other hand, the Petri net model, like other workflow engines, can be designed to run in a tightly restrictive mode, with the additional asset of the explicit determination of the workable sequences and of the branching points open to the user’s choice – which strongly suggests that it is particularly qualified for the dramatization of interactive narratives. In this paper, we argue that the two models are complementary to each other, to the point that the Petri net representation can be generated from the situation calculus model, over which an execution method, analogous to the standard token-based Petri net method, can be operated. 3.2 Deriving a Petri Net from a Situation Calculus Model In the semi-formal terminology employed in this paper, a Petri net is a graph with two kinds of nodes: places (round nodes, either empty or containing exactly one token) and transitions (square nodes representing operations).2 We define a Petri net edge as triple (Op1 , Pn, Op2 ), with two (operational) transition nodes (Op1 and Op2 ) and an intervening place node (Pn). Petri net edges are positioned so as to express a partial ordering in the execution of events, which may follow each other in linear sequences, possibly branching to form and-forks, or-forks, and-joins, and or-joins. We shall also consider a simple case of backward loop, allowing to return to a previous position and try different branching options. Our approach to generate a Petri net model is based on the observation that the ordering requirements of a Petri net can be derived from the situation calculus model. The first basic consideration is that there exists an edge connecting (through a Pn node) Op1 and Op2 if the post-conditions of Op1 have a non-empty intersection with the preconditions of Op2 . There is also an edge from Op1 to Op2 if some post-condition of Op1 cancels some post-condition of Op2 (thus causing a backward loop, whereby Op2 can be retried). Forks occur when there are edges leading from an Op node into two or more nodes Op1 , Op2 , …, Opn . A fork is an or-fork if Op1 , Op2 , …, Opn contain either incompatible pre-conditions or redundant post-conditions. Incompatibility typically results from 2 A comprehensive and still useful classic survey of the Petri net formalism is provided in [21].

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

67

conflicting value comparisons as well as from logic opposition (P vs. not P). In addition, we consider any pair Opi and Opj incompatible if the execution of one of them would be rendered impossible by the execution of the other, which might happen if the preconditions of one of these operations require P (or not P), whereas the post-conditions of the other produce the deletion of P (or, respectively, the addition of P). A fork is an and-fork if none of these situations holds. Joins occur when there are edges from two or more nodes Op1 , Op2 , …, Opn into a single node Op. A join is an or-join if Op1 , Op2 , …, Opn contain redundant postconditions, or post-conditions that cancel the post-conditions of Op (which induces a backward loop enabling Op to be retried). A join is an and-join if none of these situations holds. By thus considering the presence of edges, as well forks and joins, we have a method to automatically generate a clausal representation of the Petri net corresponding to a given situation calculus specification. In order to illustrate the results of our method, we shall return to the trial by combat example described in the previous section. By applying our method, we obtain the clausal representation below describing the edges of the Petri net. The one-letter operation labels provide a useful abbreviation, commonly employed in process mining [1] to represent transactions as parameter-less traces. (Ø, start, a:accuse(a, d, o)) (a:accuse(a, d, o), s(1), b:enter_worthy_defender(k, d, o)) (a:accuse(a, d, o), s(1), c:enter_beginner_defender(k, d, o)) (a:accuse(a, d, o), s(2), d:enter_challenger(a, d, o)) (b:enter_worthy_defender(k, d, o), s(3), e:combat(a, k, d, o, v)) (c:enter_beginner_defender(k, d, o), s(3), e:combat(a, k, d, o, v)) (d:enter_challenger(a, d, o), s(4), e:combat(a, k, d, o, v)) (e:combat(a, k, d, o, v), s(5), f:reinitiate_trial(a, k, d, o, v)) (e:combat(a, k, d, o, v), s(5), g:vindicate(d, o)) (e:combat(a, k, d, o, v), s(5), h:condemn(d, o)) (f:reinitiate_trial(a, k, d, o, v), s(1), b:enter_worthy_defender(k, d, o)) (f:reinitiate_trial(a, k, d, o, v), s(1), c:enter_beginner_defender(k,d,o)) (f:reinitiate_trial(a, k, d, o, v), s(2), d:enter_challenger(a, d, o)) (g:vindicate(d, o), end, Ø) (h:condemn(d, o), end, Ø)

The detected cases of or-forks and or-joins are listed below: Or-forks: accuse(a, d, o) → enter_worthy_defender(k, d, o), enter_beginner_defender(k, d, o) combat(a, k, d, o, v) → condemn(d, o), vindicate(d, o), reinitiate_trial(a, k, d, o, v) reinitiate_trial(a, k, d, o, v) → enter_worthy_defender(k, d, o), enter_beginner_defender(k, d, o) Or-joins: combat(a, k, d, o, v) ← enter_worthy_defender(k, d, o), enter_beginner_defender(k, d, o) enter_beginner_defender(k, d, o) ← accuse(a, d, o), reinitiate_trial(a, k, d, o, v) enter_worthy_defender(k, d, o) ← accuse(a, d, o) reinitiate_trial(a, k, d, o, v) enter_challenger(a, d, o) ← accuse(a, d, o) reinitiate_trial(a, k, d, o, v)

68

E. S. de Lima et al.

Once the clausal representation is generated, a visual representation of the Petri net can be created, as shown in Fig. 1.

s(2)

d

s(4)

enter_challenger

h start

a

c

e

accuse

enter_beginner _defender

combat

s(1)

s(3)

s(5)

condemn

end

g vindicate

b

reiniate_trial

enter_worthy _defender

f

Fig. 1. Petri net drawn from the clausal representation derived from the situation calculus model.

Operations coming from and-forks can be executed in any order. In addition, they might be executed in parallel to simulate narrative events taking place at the same time. As a preliminary consideration, note that, by construction, or-forks stem from place nodes, and or-joins always converge to a single place node. Since place nodes can contain at most one token, or-type branching is restricted – as should be expected – to the selection of a single option. In contrast, and-forks stem from operation nodes, and and-joins converge to an operation node. Differently from place nodes, operation nodes are able to emit tokens to all outgoing place nodes (one token for each). The generated Petri net is ready to be executed in order to establish the plot for an interactive narrative. According to the standard token-based Petri net process, executing a Petri net begins by placing a token in the start place node. The place node is then activated which signifies that the token is consumed, and the single operation node attached to the start node is enabled. In the next steps, successive place nodes are activated after receiving tokens from enabled nodes, and some operation node to which all incoming place nodes are active (i.e., contain a token) is chosen to be enabled. The process ends when some operation node connected to the end place node is reached. An interactive trace-generation program that allows users to traverse through the Petri net generated for the trial by combat is available online at: http://www.icad.puc-rio.br/~logtell/petri-net/trial-by-combat/.

4 Interactive Storytelling Application In order to validate the applicability of our method, we implemented a full interactive storytelling system capable of representing the generated Petri net models using 2D graphics and animations. 4.1 System Architecture The architecture of our interactive storytelling system is based on a client-server model (Fig. 2), where the server is responsible for the generation of the plot (Petri net model) and

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

69

the client handles the dramatization of the story. On the server-side, the Network Manager receives plot requests from clients and uses the Prolog implementation described in the previous sections to generate a Petri net, which is then sent to clients for dramatization. On the client-side, the Drama Manager interprets and controls the execution of the Petri net by sending action requests to virtual Actors. The process of composing scenes for dramatization (i.e., selecting the Actors and Locations to show) is performed by the Scene Composer, which is constantly being informed by the Drama Manager about the type of scene being dramatized. User interaction is handled by the Interaction App module, which is implemented as a mobile app that uses a Convolutional Neural Network classifier to identify handdraw sketches (see [20] for more details about the sketch recognition process). Once a sketch is recognized, its identification class is sent to the Interaction Server through a TCP/IP network message. The Interaction Server module is responsible for receiving and interpreting the sketch classes sent by clients. Two interaction modes are supported: (1) single user mode, in which the first valid user sketch received by the system is immediately used as the interference choice to be incorporated into the story; and (2) voting mode, in which the Interaction Server collects all users’ sketches during a certain time and then selects one through a voting process. Server

Client

Plot Generator

plot request

Network Manager

petri net

plot request

petri net

user interacons

Drama Manager

acon requests

user sketches

Actors

scene Scene Composer

Prolog

Interacon Server

Dramazaon

Locaons

Interacon App

Fig. 2. Architecture of our interactive storytelling system.

Multiple programming languages were used in the implementation of our interactive storytelling system. As described in the previous sections, the process of generating plots in the Petri net model is implemented in Prolog. However, the Plot Generator also includes an additional module called Network Manager, which is implemented in C# and provides network communication capabilities to the system, allowing us to implement

70

E. S. de Lima et al.

the plot generation process as a service provided by a network server. On the clientside, the dramatization system is implemented in Lua3 using the Löve 2D framework,4 which provides the graphical functionalities needed to create visual representations for the story. The interaction process is implemented in Java as an Android app, which communicates with a Web service implemented in PHP. See [12] for more details about the design of the sketch-based interaction system. The Petri net representation of the plot created by the Plot Generator consists of a directed graph G = (V, E), where V is a set of nodes {v1 , v2 , …, vn } and E is a set of edges {ei = (vi , vj ), …, em = (vk , vw )}. Each node vi is a pair (id i , evi ), where id i is a unique name that identifies the node vi and evi is an event description in a predicate format for transition nodes (e.g., accuse(a,d,o)), or the constant nil for place nodes (as described in Sect. 3.2, places are nodes that can contain tokens and transitions are nodes that represent operations). When encoding the Petri net to be sent to the dramatization system, the graph is simplified as a set of edges, where each edge is represented in the format [id i : evi , id j : evj ]. For example, the initial edges of the Petri net generated for the trial by combat (connecting nodes start, a, s(1), and s(2), as illustrated in Fig. 1), can be described as: [start:nil, a:accuse(a, d, o)] [a:accuse(a, d, o), s(1):nil] [a:accuse(a, d, o), s(2):nil]

4.2 Interactive Dramatization The process of dramatizing the Petri net representation of the plot involves a simple stepwise algorithm that controls the execution of the story by updating a list of active events according to a standard token-based execution approach. As described in Algorithm 1, function Execute-PetriNet-Step receives by parameter a Petri net PN and a list C with the nodes that were executed in the previous step of the algorithm (for the first step: C={start}). The algorithm performs all the operations to activate place nodes and transition nodes for a single iteration of the execution process. The narrative events associated with activated transition nodes are added to set A, which is returned when the execution of the iteration ends. The set of narrative events returned by a single call of function Execute-PetriNet-Step represents the parallel events that take place during a certain point of the narrative. When the dramatization of these events ends, function Execute-PetriNet-Step can be called again to obtain the next narrative events for dramatization. If an empty set is returned, the narrative ends.

3 Lua is a well-known programming language developed at the Pontifical Catholic University of

Rio de Janeiro, Brazil (http://www.lua.org/). 4 https://love2d.org/.

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

71

Algorithm 1. Petri net execution algorithm. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32.

function Execute-PetriNet-Step(PN, C) A = ∅; for each node V in C do if PN[V] is a PLACE then N = number of edges in PN[V]; if N is greater than 0 then if N is 1 then S = first edge in PN[V]; else S = get selected edge from PN[V] based on user interaction; end TA = get number of tokens available in parent nodes of PN[S]; TN = get indegree of PN[S]; if TA is greater or equal than TN then Consume TN tokens from the parent nodes of PN[S]; Add S to A; end end else if PN[V] is a TRANSITION then for each edge E in PN[V] do if PN[E] is a PLACE then Add a token to place PN[E]; L = Execute-PetriNet-Step(PN, {E}); for each node W in L do Add W to A; end end end end end return A; end

All the assets used for dramatization (e.g., character animations, background images, and audio files) are defined in a library manually constructed for the domain of a specific story. The context library is a 5-tuple L = (γ, α, β, δ, π), where: • γ is a set that defines the actors of the story. Each actor has a name and a set of actions, which are represented by animations in a sprite sheet format; • α defines the locations of the story. Besides associating each location with a background image and a soundtrack, it also defines a set of waypoints where actors can be placed during the scene composition process; • β defines the characters’ dialogs (text and audio); • δ is a set that defines the interaction points of the story. Each interaction point is associated with a set of interactive objects, which are represented by the classes of sketches that can be used by users to interact at each interaction point. The interaction points also include a set of instructions to guide the user during the interaction; • π establishes values for the variables present in the events of the Petri net. In our implementation, the context library is defined in an XML file. The library used for the trial by combat example is available at: http://www.icad.puc-rio.br/~logtell/ petri-net/context-trial-by-combat.xml. During the dramatization of the story, our system generates 2D animations in real time according to the actions performed by the virtual actors. An automatic virtual camera

72

E. S. de Lima et al.

maintains the active actors always centered in the image frame while they move around the virtual world. When more than one actor is involved in the action, the camera will target the center of the scene, which is calculated based on the positions of all characters that are participating in the event. User interaction occurs at or-fork nodes of the Petri net. When a node of this type is activated, users are instructed by the virtual characters to interact by drawing specific objects in the interaction app. The instructions are defined in the context library and comprise a set of phases (text and audio), which are repeated until the user draws a valid object (single user interaction mode) or during a certain time frame (voting interaction mode). When the user’s choice is identified, the corresponding transition node is selected to be activated (as indicated in line 10 of Algorithm 1). An example of a user interaction moment for the trial by combat is illustrated in Fig. 3, showing the user’s decision whether to help Gawain or Perceval, in the combat by drawing a spear or a sword (i.e., the weapons used by each character). A complete video demonstration of the trial by combat example is available at: https://www.youtube.com/watch?v=qI2TeBrhycc.

(a)

(b)

Fig. 3. User interaction moment in the trial by combat: (a) shows the dramatization system instructing the user to draw a spear to assist Gawain or a sword to help Perceval; and (b) shows that the user chose to draw a sword in the interaction app.

5 Concluding Remarks We claim that our research thus far has already revealed the advantages of the complementary use of situation calculus and Petri nets. The situation calculus model is most convenient to start with, allowing to investigate through a planning algorithm the appropriateness of the initial specification. On the other hand, mapping the situation calculus schemas into the graphic structure of a Petri net permits the identification of the points where the narrative process proceeds along branching sequences, so as to recognize and explicitly annotate the occurrence

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

73

of forks, joins and loops. This kind of information is most helpful to guide interactive plot generation/dramatization and is indispensable if dramatization is done by putting together video-recorded sequences [18], given that scene transition often poses nontrivial adjustment problems that cannot be left to be solved at runtime. In addition, one must recall the relevance of this method to game design [16], where multiple-ending and branching path storytelling mark increasingly advanced stages in the interactivity spectrum, with remarkable examples, such as the Mass Effect trilogy (BioWare, 2007–2012) and The Witcher trilogy (CD Projekt RED, 2007–2015). Due to their predictability, handcrafted branching narrative structures are still dominant in the game industry. However, we believe that more open approaches to interactive storytelling, such as our method, can expand the boundaries of game narratives towards new forms of interactive experiences. The situation calculus used at the specification stage of our approach gracefully deals with unpredictability because the complete sequence of outcomes is not explicit in the set of operators but can be easily verified. Much work, however, remains to be done. As a proof of concept, we initiated this project working upon an oversimplified example. Accordingly, we do not claim that the current prototype can handle all problems associated with more complex applications and intricate Petri net schemes. For instance, Petri net loops caused by iterative actions have not been considered. Also, we could enrich the information kept at each Petri net node by collecting user behavior data during a run and analyzing them to regulate the branching options. Another essential future investigation is to address non-deterministic events, i.e., events that can have more than one outcome. Finally, we also plan to explore authoring systems to support story writers, and to conduct comprehensive user satisfaction tests involving writers (i.e., authors) and players (i.e., co-authors) in future works. Acknowledgements. We want to thank CNPq (National Council for Scientific and Technological Development) and FINEP (Funding Agency for Studies and Projects), which belong to the Ministry of Science, Technology, and Innovation of Brazil, for the financial support.

References 1. Aalst, W.V.D.: Process mining. Commun. ACM 55(8), 76–83 (2012). https://doi.org/10.1145/ 2240236.2240257 2. Agostinelli, S., Maggi, F.M., Marrella, A., Mecella, M.: Verifying petri net-based process models using automated planning. In: Proceedings of the 2019 IEEE 23rd International Enterprise Distributed Object Computing Workshop (EDOCW), pp. 44–53. IEEE Press, New York (2019). https://doi.org/10.1109/EDOCW.2019.00021 3. Balas, D., Brom, C., Abonyi, A., Gemrot, J.: Hierarchical petri nets for story plots featuring virtual humans. In: Proceedings of the Fourth AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE’08), pp. 2–9. AAAI Press, Menlo Park (2008) 4. Brom, C., Abonyi, A.: Petri-nets for game plot. In: Proceedings of AISB Artificial Intelligence and Simulation Behaviour Convention 3, pp. 6–13 (2006) 5. Ciarlini, A.E.M., Barbosa, S.D.J., Casanova, M.A., Furtado, A.L.: Event relations in planbased plot composition. Computers in Entertainment 7(4), 55 (2009). https://doi.org/10.1145/ 1658866.1658874

74

E. S. de Lima et al.

6. Ciarlini, A.E.M., Casanova, M.A., Furtado, A.L., Veloso, P.A.S.: Modeling interactive storytelling genres as application domains. J. Intelligent Inf. Systems 35(3), 347–381 (2010). https://doi.org/10.1007/s10844-009-0108-5 7. Ciarlini, A.E.M., Pozzer, C.T., Furtado, A.L., Feijó, B.: A logic-based tool for interactive generation and dramatization of stories. In: Proceedings of the International Conference on Advances in Computer Entertainment Technology (ACE 2005), pp. 133–140. ACM Press, New York (2005). https://doi.org/10.1145/1178477.1178495 8. Eco, U.: The Open Work. Harvard University Press, Cambridge (1989) 9. El-Sattar, H.K.H.A.: A new framework for plot-based interactive storytelling generation. In: Proceedings of the 2008 Fifth International Conference on Computer Graphics, Imaging and Visualisation, pp. 317–322. IEEE Press, New York (2008). https://doi.org/10.1109/CGIV.200 8.50 10. Fernandes, A., Ciarlini, A.E.M., Furtado, A.L., Hinchey, M.G., Casanova, M.A., Breitman, K.K.: Adding flexibility to workflows through incremental planning. Innovations Syst. Softw. Eng. 3(4), 291–302 (2007). https://doi.org/10.1007/s11334-007-0035-y 11. Fikes, R.E., Nilsson, N.J.: A new approach to the application of theorem proving to problem solving. Artif. Intell. 2(3–4), 189–208 (1971). https://doi.org/10.1016/0004-3702(71)90010-5 12. Gheno, F., Lima, E.S.: História viva: a sketch-based interactive storytelling system. In: Proceedings of the XX Brazilian Symposium on Computer Games and Digital Entertainment (SBGames 2021), pp. 116–125. SBC, Porto Alegre (2021) 13. Hickmott, S., Rintanen, J., Thiebaux, S., White, L.: Planning via petri net unfolding. In: Proceedings of the 20th International Joint Conference on Artificial intelligence, pp. 1904– 1911. AAAI Press, Menlo Park (2007) 14. Kindler, E., Rubin, V., Schäfer, W.: Process mining and petri net synthesis. In: Eder, J., Dustdar, S. (eds.) BPM 2006. LNCS, vol. 4103, pp. 105–116. Springer, Heidelberg (2006). https://doi.org/10.1007/11837862_12 15. Kowalski, R., Sadri, F.: Reconciling the event calculus with the situation calculus. J. Logic Programming 31(1–3), 39–58 (1997) 16. Lebowitz, J., Klug, C.: Interactive Storytelling for Video Games: A Player-Centered Approach to Creating Memorable Characters and Stories. Focal Press, Waltham (2011) 17. Lee, Y.-S., Cho, S.-B.: Dynamic quest plot generation using Petri net planning. In: Proceedings of the Workshop at SIGGRAPH Asia (WASA ‘12), pp. 47–52. ACM Press, New York (2012). https://doi.org/10.1145/2425296.2425304 18. de Lima, E.S., Feijó, B., Furtado, A.L.: Video-based interactive storytelling using real-time video compositing techniques. Multimedia Tools Appl. 77(2), 2333–2357 (2017). https://doi. org/10.1007/s11042-017-4423-5 19. Lima, E.S., Furtado, A.L., Feijó, B., Casanova, M.A.: A note on process modelling: combining situation calculus with petri nets. Technical Report 01/2022, Department of Informatics, PUC-RIO, Rio de Janeiro (2022). https://doi.org/10.17771/PUCRio.DImcc.59758 20. Lima, E.S., Gheno, F., Viseu, A.: Sketch-based interaction for planning-based interactive storytelling. In: Proceedings of the XIX Brazilian Symposium on Computer Games and Digital Entertainment (SBGames 2020), pp. 348–356. IEEE Press, New York (2020). https:// doi.org/10.1109/SBGames51465.2020.00029 21. Peterson, J.L.: Petri nets. ACM Comput. Surv. 9(3), 223–252 (1977). https://doi.org/10.1145/ 356698.356702 22. Peterson, J.L.: Petri Net Theory and the Modeling of Systems. Prentice Hall, Upper Saddle River (1981) 23. Riedl, M., Li, B., Ai, H., Ram, A.: Robust and authorable multiplayer storytelling experiences. In: Proceedings of the Seventh AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE’11), pp. 189–194. AAAI Press, Menlo Park (2011)

Plot Composition by Mapping Situation Calculus Schemas into Petri Net

75

24. Tan, X.: SCOPE: A situation calculus ontology of Petri Nets. In: Proceedings of 6th International Conference of Formal Ontology in Information Systems, Toronto, Canada, pp. 227–240 (2010) 25. Ziparo, V.A., Iocchi, L., Nardi, D., Palamara, P.F., Costelha, H.: Petri net plans - a formal model for representation and execution of multi-robot plans. In: Proceedings of the 7th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2008), pp. 79–86 (2008)

Upward Influence Tactics: Playful Virtual Reality Approach for Analysing Human Multi-robot Interaction Cornelia Gerdenitsch1(B) , Matthias Weinhofer2 , Jaison Puthenkalam1 , and Simone Kriglstein1,2,3 1

2

AIT Austrian Institute of Technology GmbH, Vienna, Austria {cornelia.gerdenitsch,jaison.puthenkalam}@ait.ac.at Faculty of Computer Science, University of Vienna, Vienna, Austria [email protected] 3 Masaryk University, Brno, Czech Republic [email protected]

Abstract. The interest, the potential, and also the technical development in artificial intelligence assistants shows us that these will play an essential role in the future of work. Exploring the interaction and communication between human and artificial intelligence (AI) assistants forms the basis for the development of trustworthy and meaningful AI-based systems. In this paper we focused on the question how humans react to AI - more precisely, AI gents as robots - that act to influence human behavior and emotions by using two upward influencing tactics: Ingratiating and Blocking. For this purpose, we developed a playful virtual reality approach that creates a leader-subordinate relationship between humans and the AI agents in a factory environment. We explore how humans react to those agents. Among other things, we found that behaviors that are seen as likable in humans are perceived as distracting in robots (e.g., compliments used by the ingratiating tactic). Further, robots were perceived as a group and not as individuals. Our findings showed us directions and open questions which need to be investigated in future work investigating human-multi-robot interaction at the workplace. Keywords: Virtual reality Human robot interaction

1

· Robot · Leadership · Influence tactics ·

Introduction

Let’s imagine a fully automated production line in which a large number of industrial robots handle all manual tasks. The role of the human is to control and monitor the machines. In our specific scenario, a single human works with multiple robots and has authority/power. If we now compare this human multi-robot c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 76–88, 2022. https://doi.org/10.1007/978-3-031-20212-4_6

Human Multi-robot Interaction in Virtual Reality

77

system with the relationships between humans in organizations, this corresponds to a leadership situation; the human is the leader and the robots are the subordinates. We now ask how the use of influence tactics, such as those used by employees towards managers, affects humans when used by robots. To address this question, we simulated this scenario in a virtual reality (VR) environment. In this paper, we present our findings from this VR-based factory environment that creates a leader-subordinate relationship consisting of a human as the leader and multiple robots as the subordinates. Using artificial intelligence, we equipped the robots with upward influence tactics - tactics used by employees to steer the leader in a particular direction [5]. These involve the ingratiation tactic - showing approval or praise - and the blocking tactic - threatening to work more slowly or become uncooperative. We chose a playful approach with challenges and earnable ‘credits’ to support the user’s interaction with the robots over time. The use of a virtual environment also allows us to safely simulate the factory floor and be free in the design of the robots. An evaluation was conducted to obtain feedback on the designed VR environment and to gain initial insight into how users respond to the robots.

2

Related Work

The way interactions between humans and robots take place in industrial workplaces is described as coexistence (working side by side), cooperation (working on the same product on different tasks) or collaboration (working together on one task) [1]. In this context, robots and humans are work colleagues operating on a similar hierarchical level. However, when authority and power is given to the human who is tasked with instructing and supervising the robots, this is similar to a leader-subordinate relationship; the leader is a human and the subordinates are robots. Although most research on leadership focuses on the behavior of leaders (i.e., leadership styles; [25]) and how they influence their subordinates, leadership is not an exclusively one-way process. Rather, leadership can be from both sides [28]. Subordinates can influence leaders in the form of upward influence tactics [12,20,24,32]. By bringing the interpersonal influence theory [16] to the organizational setting, Kipnis and colleagues [23] identified a categorization of eight influence tactics including assertiveness, ingratiation, rationality, sanctions, exchange, upward appeals, blocking and coalitions. Since than these tactics have been extended by several authors including [15,32]. The use of a specific tactic depends on a number of contextual and individual factors. Individuals differ for instance in the influence tactic they apply. For instance, one may use self-promotion to obtain a job promotion another one ingratiation [20]. Tactics can be separated into soft - such as ingratiation - and hard - such as assertiveness - ones [14]. For this research we selected one soft and one hard tactic. The tactic Ingratiation is shown when employees use compliments to flatter their leaders in order to be treated better or to gain a more positive reputation [19,20,32]. Using ingratiation as influence tactic has a positive effect

78

C. Gerdenitsch et al.

on performance assessment and extrinsic success [20]. Blocking, on the other hand, is formulating a threat of becoming uncooperative if demands are not met, such as working more slowly or ignoring the instructor [5]. Upward influence tactics have been shown not only to affect performance on the job [20] but also to affect the quality of the relationship between leaders and subordinates. In particular, it has been shown that upward influence tactics affect the extent to which leaders favor some subordinates over others in their decisions, as the tactics affect mutual respect [9]. Also, Falbe and Yukl [13], demonstrated that a single soft tactic is more effective than a singly hard tactic. The interaction between humans and robots have been researched for instance in terms of the appearance of robots [18] and their nonverbal [30] or verbal communication [29]. Further it has been highlighted that the perceived formal authority of a robot influences human behavior. Particularly, it has been shown that robots that are perceived as peers are more persuasive than those with formal authority [31]. Recent research investigates the interaction between humans and multiple robots that are either represented by teams [6,27] or swarms [22]. In this context, the design of the motions of the robot swarm [22] and the interface design of the remote control through which the robot swarm can be instructed [7,26] has been explored so far. Within this work, we contribute to existing research by exploring how humans experience the interaction with multi-robots when these show behavior that is intended to influence them. In this context, we use artificial intelligence to model autonomous behavior in a way that robots can act independently to some extend.

3

Approach

A VR factory environment was developed using Unreal Engine 4 [11]. In this virtual environment the user acts as a leader of factory workers, that are represented by virtual robot-like characters. In the following sections we describe the robot avatar, the virtual environment and interaction modes, the task of the game, and the robot behavior. 3.1

Robot Avatar

We used Blender 2.8 [4] to create and animate the robot avatar. Based on previous evidence and animation guidelines [33] initial designs were created. A pretest was conducted with 22 participants to gain first feedback from users (Fig. 1 A to E). Within an online survey we asked those users to assess all designs regarding likability, reliability, trustworthiness, competence, intelligence, perceived threat and innocence on a Likert scale from 1 (lowest) to 5 (highest). Participants were then asked to select their most preferred design and rate which variant of the head, body, and eyes design they liked best. The feedback from the participants indicated a preference to designs A and C. We decided to implement design A, since the participants perceived this design as more likeable, trustworthy and

Human Multi-robot Interaction in Virtual Reality

79

Fig. 1. Initial robot designs (A-E) from two perspectives (front and side), that have been subject to pre-testing.

innocent compared to design C. We argue that these properties may help the bonding between the user and robots. The final robot design comprised humanoid characteristics including a face, legs and arms (see Fig. 2). In particular, the robot avatar is a white cube with a display which shows two eyes. The arms and legs are white cones. Humanoid robots are preferred in interactions as they garner more trust [8]. Further, it was shown that personality traits were more likely to be attributed to them compared to non-humanoid designs [8]. To represent emotions through facial expressions, the robot was designed with a large head and eyes (see Fig. 3). Previous research highlight facial expressions as a essential feature of a communicative robot [2,21]. To make robots distinguishable in the virtual environment, colored armlets (blue, red, pink, yellow, green) and numbers (1 to 5) on the backside of the robot are used.

Fig. 2. Final design of the robot from three perspectives (front, side, back).

3.2

Virtual Environment and Interaction Modes

The factory environment is designed to be open and on a single level. Two rooms were designed: a leader’s office and a factory floor (see Fig. 4). A glass wall is located between the two spaces, through which the robots movements can be observed. The leader’s office serves to represent the player’s status and authority as a manager. Artifacts such as a large desk and office decorations

80

C. Gerdenitsch et al.

Fig. 3. Three emotional expressions: angry, happy, neutral. Right image: A robot transporting an item on its head, which was picked up from the material supply machine.

(a plant, an abstract painting, and a cup) are used to support this. By moving the head, the player could overlook the entire factory floor. Goods were produced on the factory floor. Five robots operated three machines that produced one of three different goods: tetrahedrons, cubes and dodecahedrons. Each good was assigned to a machine, graphically represented at the top of the machines. A good can only be produced if the corresponding machine has at least one source material stored. Thus, in a first step, materials must be picked up from the material supply by a robot and brought to the machine. The status of production is displayed on the machine by a progress bar. When a product is finished, it appears at the left side of the machine. A robot then needs to pick the good up and bring it to a fume hood. An image above their heads indicate which good a robot is carrying at the moment. Interaction in the virtual environment is realized through point and click operations via the VR-controllers. The interaction with the robots is realized through a terminal placed on the leaders desk (see Fig. 5). Four pages are implemented. On the first one the user finds a summary about the actual progress; how many goods have been produced so far. The second one lists all robots. After selecting one robot a third page automatically appears. Tasks can then be assigned to the selected robot. Tasks are i) operating on one of the machines, ii) supplying the machine with materials, iii) collecting all produced goods (robots autonomously seek out finished goods and carry them to the flume hub), and iv) calling robot to the office. Finally, the fourth page summarises the robots’ communication. When a robot is attending the office, it formulates requests and thoughts which are depicted at this terminal page.

Human Multi-robot Interaction in Virtual Reality

81

Fig. 4. Examples from the environment. Left: The place from the player (desk, a display, coffee cup and a plant) and the view to the factory room with the five robots. Robot 2 is operating the machine producing cubes, and robot 5 is operating the machine producing dodecahedrons. Right: A conversation between the player and a robot.

Fig. 5. Terminal.

3.3

Task of the Game

The task of the game is the production of a certain amount of goods within a given time limit. Both aspects can be configured at the beginning of the game, with a default setting of seven tetrahedrons, seven cubes, and seven dodecahedrons to be produced within ten minutes. In order to manage this challenge, the player has to assign robots to specific operations. Credits are awarded to the player whenever a produced good was ‘delivered’ (sucked into the fume hood). The current status of produced goods and the time remaining is shown at the terminal.

82

C. Gerdenitsch et al.

The game session was split in two phases, each of which accounted for half of the defined time. In phase one, the robots follow orders from the players without interacting. In phase two, robots start interacting with the player. They walk in front of the glass wall and wave to the player as a request to come into the office. The player can then decide to let the robot in or not. When the robot is let in, it moves in front of the desk and stands on a marked spot, which then moves up and brings the robot up to desk height. Now the robot can express a request (e.g., ‘I want another task’) which is shown at the terminal. The player can then comply with the request (e.g. assigning a different task to the robot) or not. Then the robot leaves the office and continues work, with it’s behavior adapted based on the player’s reaction. 3.4

Robot Behavior

Robots are equipped with one of two types of behavior - Ingratiating or Blocking - which can be selected at the beginning of the game. We used a behaviour tree system provided by Unreal Engine 4 [10] to model this behavior. This tree was supplemented by additional sub-behaviour trees that would account for autonomous behaviour. Further the behaviour trees were supplemented with the ability to cycle through a variety of facial expressions for the robot, both randomly and context sensitive. Robots are able to formulate ‘requests’ and ‘thoughts’. Requests concern the task they have been assigned to, and want to have changed (e.g. ‘I would like to produce X instead of Y’). Thoughts are represented by compliments in the ingratiating group and threats (of deliberately worse performance if requests are not met) in the blocking group. The sequence and combination of requests and thoughts followed a predefined repeating storyboard. Robots showing the ingratiating behavior follow a repeating four-step pattern of thought → thought → request → both (thought and request). Thus, in 75% of the contacts, ingratiating robots formulate thoughts, while in 50% an actual request is formulated. This specific pattern was used in order to establish these robots as being more interested in positive rapport (communicated via thoughts) rather than making demands (communicated via requests). Inversely, robots that are equipped with the blocking behavior follow a cycle with a strong focus on demands: both (thought and request) → request → both → request. Thus, in 50% of the contacts robots formulate thoughts, while in 100% requests are formulated. When these requests were not met, they had a random chance of displaying uncooperative behaviours, in particular laying down their work or slowing down work speed. Every 30–40 seconds one of the robots attempt to make contact with the player in a predefined order (robot 1, robot 3, robot 5, robot 2, and finally robot 4). Thus, players would thus receive about six autonomous contact attempts within five minutes (default setting of phase 2). Further, the interaction cycle would be completed at least once.

Human Multi-robot Interaction in Virtual Reality

4 4.1

83

Evaluation Method

We conducted a study with ten participants aged between 25 and 60. Participants were randomly assigned to one of the two influence tactics. In the Ingratiating condition there were three female and two male participants, while in the Blocking condition there were three male and two female participants. All participants were able to familiarize themselves with the environment and managed the game and inputs without any major problems. The study lasted about 30 min and was organized in four steps: First, participants were provided with a written instruction with a possibility to discuss technical questions afterwards. After that, the VR glasses were put on and the participants could explore the environment (without robots) and get used to the controllers. Then the gameplay started, with a time limit was set to ten minutes. During gameplay, participants were observed and their comments were noted. Since the game action was simultaneously displayed on a screen connected to the computer running the virtual environment, observation was possible. At the end, we conducted an interview and asked the following questions: Please describe the assignment/job you were given, How did you perceive your role as leaders?, How did you perceive the robots?, What influenced your decisions to grant or dismiss a robot’s request?, How do you describe working with the robots?, What strategy did you use to complete the task?, Can you describe what factors led to your success or failure?, and What did playing the game feel like from a technical/haptic standpoint? 4.2

Findings

At the beginning participants started to establish a game strategy to meet the challenge. Three strategies were observed: In the “one type of goods at a time” strategy, two robots are assigned to a machine for production and supply and another one to goods collection. No adaptations are made, with the purpose of finishing the production of the first good first before continuing with the other ones. Four users followed this strategy, two of which expressed boredom towards the end of the first phase. Another four users employed a “parallel processing” strategy. Here, one robot is assigned to each of the three machines. Another robot has to collect the produced goods and bring them to the fume hood, and the last one supplies all machines with needed material. Finally, in the chaotic approach, the assignments of robots are changed constantly with no discernible reasoning. Two participants used this chaotic approach. One observation across all participants was that users are activated and excited during the gameplay. This activation further increases in the second phase once the robots start interacting. In this phase, some participants also expressed irritation with these interactions interfering with their strategy. None of the participants achieved the goal of 7 × 3 collected goods within ten minutes. Participants felt that the failure was their own fault, as they failed

84

C. Gerdenitsch et al.

to properly manage the robots which resulted in a lack of organisation. At the same time, some participants saw the robots as capable of the necessary agency. In particular in the blocking condition, two users mentioned to have completely relinquished control to the robots at some point, granting every request thereafter. The reason given was that they thought that the robots would know best what steps had to be taken to manage the factory successfully. Regarding the robots’ appearance, all participants reacted positively to the robots’ design and their animations, with nine stating that they perceived their cuteness as appealing and four commenting that the animations made the robots likeable. The greeting/waving animation displayed by the robots when entering the office was perceived as a polite. In terms of the robots’ behavior, nine participants expressed a preference for the robots to be more autonomous and require less direct instruction. Also the randomly displayed emotional expressions have been mentioned several times. Participants mentioned feeling insecure about their own behavior and performance which they perceived to possibly have caused these emotions. For instance, participants formulated worry about having done something incorrectly or performing poorly as a supervisor when a robot displayed an angry or unsatisfied face. One observation during the game was that participants did not differentiate between the individual robots - instead robots were perceived as a unit. Numbers on the back and colored armlets were used to differentiate between the robots. However, these markers were only used for giving commands and seems to have not lead to participants differentiating between the robots in other aspects - for instance by having a favor for one or another. This also led to the consequence that actions of a single robot ‘affected’ all the other ones. Ingratiating Influence Tactic. Participants from the Ingratiating group accepted on average 87.5% of the robot requests. The most often given reasons for this were that the requests fit the game strategy anyway, as well as that there was no obvious reason to refuse the request. Once the robots started acting autonomously and interacting with the player, participants showed stress and hectic behavior. This disrupted the game strategies that had been built up in the first phase, and participants then tended to focus on interacting with the robots and to process them as quickly as possible. When asked, How did you perceive your role as a leader? participants indicated that their role was relatively low-stress and only became slightly more stressful when the robots began to act autonomously. Two participants in the group were amused or felt complimented by the positive messages from the robots. However, four out of five expressed dissatisfaction that the robots left their work to give encouragement. Participants mentioned that the robots should only interrupt their work when they were not assigned a task or after all tasks were completed. The expressed dissatisfaction with the robots increased with each message that did not contain a specific request. Blocking Influence Tactic. Participants from the Blocking group accepted on average 81% of robot’s requests. Two players who approved all of the requests

Human Multi-robot Interaction in Virtual Reality

85

indicated that they assumed that the robots would likely know which actions were best to achieve the stated goals. Therefore, they left the decision-making power to the robots and approved each request from the robots. In response to the question How did you perceive your role as a supervisor? all five participants indicated that they found their role stressful. It is worth noting that users did not describe any particular phase as stressful, which distinguishes their descriptions from those of the Ingratiating group, where only the second phase was described as stressful. A further observation was that the participants did not seem to notice the robots acting on their threats if their demands were not met (e.g. by becoming slower over time). Instead participants perceived the robots to be acting in a positive and friendly way. Three participants further mentioned not wanting to disappoint the robots and wanting to be a good and friendly boss for them.

5

Discussion

In the present study we used VR to simulate a human multi-robot interaction at an industrial workplace analogous to a leader-subordinate relationship. A evaluation of the virtual simulation yielded insights that should guide further developments and research. Power and Ability: The user as a leader had the power and authority to give instructions to the robots, but preferred to have the robots do the tasks autonomously with less direct instruction. It was generally assumed that the robot has the necessary ability and can best decide for itself how the challenge can be met. This was expressed through a desire to hand over tasks to the technology. We recommend for future studies to investigate what is associated with this desire. Themes can be for instance, the perception of one’s own skills and that of the robots, issues of responsibilities or previous experiences in humanrobot interactions. Individual Robot within a Swarm: We found indications that robots are seen as a group rather than as individual entities. Robots were not afforded individuality, and actions of a single robot affect the robots as a whole. One explanation could be the design of the robots, which was not distinctive enough. Also, forming relationships with individual robots would require more time than we used in the present study. Likewise leader-subordinate relationship form over time by recurrent interactions. Future work could explore if relationships with single robots emerge when interacting with them on a daily basis. It would also be interesting to investigate design principles that support this evolving process. For instance, previous research highlight that body movement and gestures [17], and dimensions of anthropomorphism [3] convey affect. In addition, individual facial expressions (instead of random ones for all) can support individual robot bonding and discrimination. Influence Tactics: It appeared that users were irritated when the robots started to interact. The reaction to the robot behaviors showed peculiarities. We found,

86

C. Gerdenitsch et al.

for example, that behaviors that are considered likeable in humans are considered annoying in robots. Participants were irritated and, over time, annoyed when robots approached them to make friendly statements. Behaviours like compliments were perceived as distracting rather than likable. Also blocking behavior was attributed to own mistakes rather than to inappropriate behavior by the robot. Despite initial findings on how robots can exert influence, this was limited to the formulation of very simple requests and thoughts. Future research should focus more on communication and investigate different forms of it.

6

Conclusion

In this paper, we explored the interaction between humans and multiple AIdriven robots in an industrial setting. We placed participants in a VR environment in the role of a manager (in a leader role) managing a group of robots (in a subordinate role) that were equipped with upward influencing tactics. We discovered that behaviours like compliments were perceived as distracting rather than likable. Furthermore, despite differences in appearance, the robots were perceived as a group rather than individuals, and actions of one robot could influence the perception of others. These and other initial findings described in this paper showcase promising areas for future research to explore the relationship between humans and multi AI-driven robots in the context of work. Acknowledgements. This work is supported by the Austrian Research Promotion Agency (FFG) within the project “Virtual Skills Lab” (FFG No. 872573) and the project “MED1stMR” (Medical First Responder Training using a Mixed Reality Approach featuring haptic feedback for enhanced realism) funded by the H2020 program (Grant Agreement No. 101021775).

References 1. Aaltonen, I., Salmi, T., Marstio, I.: Refining levels of collaboration to support the design and evaluation of human-robot interaction in the manufacturing industry. Procedia CIRP 72, 93–98 (2018). https://doi.org/10.1016/j.procir.2018.03.214 2. Bartneck, C., Kanda, T., Mubin, O., Al Mahmud, A.: Does the design of a robot influence its animacy and perceived intelligence? Int. J. Soc. Robot. 1(2), 195–204 (2009) 3. Bartneck, C., Kuli´c, D., Croft, E., Zoghbi, S.: Measurement instruments for the anthropomorphism, animacy, likeability, perceived intelligence, and perceived safety of robots. Int. J. Social Robot. 1, 71–81 (2009). https://doi.org/10.1007/ s12369-008-0001-3 4. Blender Foundation: Blender 2.8 (2022). www.blender.org/download/releases/280 Accessed Jan 12 2022 5. Blickle, G.: Einflusstaktiken von mitarbeitern und vorgesetztenbeurteilung: Eine pr¨ adiktive feldstudie. [employee influence tactics and supervisor evaluation: a predictive field study.]. Zeitschrift f¨ ur Personalpsychologie 1 (2003). https://doi.org/ 10.1026//1617-6391.2.1.4

Human Multi-robot Interaction in Virtual Reality

87

6. Chan, W.P., et al.: Design and evaluation of an augmented reality head-mounted display interface for human robot teams collaborating in physically shared manufacturing tasks. J. Hum.-Robot Interact. (2022). https://doi.org/10.1145/3524082 7. Chen, J.Y.C., Haas, E.C., Barnes, M.J.: Human performance issues and user interface design for teleoperated robots. IEEE Transactions on Systems, Man, and Cybernetics, Part C Appl. Rev. 37(6), 1231–1245 (2007). https://doi.org/10.1109/ TSMCC.2007.905819 8. Choudhury, A., Li, H., Greene, C., Perumalla, S.: Humanoid robot-application and influence. Archives Clin. Biomed. Res. 2, 198–227 (2018). https://doi.org/10. 26502/acbr.50170059 9. Clarke, N., Alshenalfi, N., Garavan, T.: Upward influence tactics and their effects on job performance ratings and flexible working arrangements: the mediating roles of mutual recognition respect and mutual appraisal respect. Hum. Resour. Manage. 58(4), 397–416 (2019). https://doi.org/10.1002/hrm.21967 10. Epic Games Inc.: Behavior trees (2004-2022), docs.unrealengine.com/4.27/enUS/InteractiveExperiences/ArtificialIntelligence/BehaviorTrees. Accessed Jan 12 2022 11. Epic Games Inc.: Unreal engine (2004-2022). https://www.unrealengine.com/enUS Accessed Jan 12 2022 12. Epitropaki, O., Martin, R.: Transformational-transactional leadership and upward influence: the role of relative leader-member exchanges (rlmx) and perceived organizational support (pos). Leadersh. Q. 24(2), 299–315 (2013). https://doi.org/10. 1016/j.leaqua.2012.11.007 13. Falbe, C.M., Yukl, G.: Consequences for managers of using single influence tactics and combinations of tactics. Acad. Manage. J. 35(63) (1992). https://doi.org/10. 5465/256490 14. Farmer, S., Maslyn, J., Fedor, D., Goodman, J.: Putting upward influence strategies in context. J. Organ. Behav. 18, 17–42 (1997). https://doi.org/10.1002/ (SICI)1099-1379(199701)18:117::AID-JOB7853.0.CO;2-9 15. Ferris, G., Judge, T., Rowland, K., Fitzgibbons, D.: Subordinate influence and the performance rating process; test and a model. Organ. Behav. Hum. Decis. Process. 58, 101–135 (1994). https://doi.org/10.1006/obhd.1994.1030 16. Gardner, W.L., Martinko, M.J.: Impression management in organizations. J. Manag. 14(2), 321–338 (1988). https://doi.org/10.1177/014920638801400210 17. Ginevra, C., Villalba, S.D., Camurri, A.: Recognising human emotions from body movement and gesture dynamics. Affective Computing and Intelligent Interaction, pp. 71–82 (2007). https://doi.org/10.1007/978-3-540-74889-2 7arXiv:9780.20139 18. Goetz J, Kiesler S, P.A.: Matching robot appearance and behavior to tasks to improve human-robot cooperation. In: IEEE International Work Robot Human Interact Communication, vol. 11, 575–608 (2003). https://doi.org/10.1109/ ROMAN.2003.12517960 19. Gordon, R.A.: Impact of ingratiation on judgments and evaluations: a metaanalytic investigation. J. Personality Social Psychol. 71(54) (1996). https://doi. org/10.1037/0022-3514.71.1.54 20. Higgins, C.A., Judge, T.A., Ferris, G.R.: Influence tactics and work outcomes: A meta-analysis. J. Organ. Behav. 24(1), 89–106 (2003). http://www.jstor.org/ stable/4093798 21. Kao, Y.H., Wang, W.J.: Design and implementation of a family robot. In: 2015 12th International Joint Conference on Computer Science and Software Engineering (JCSSE), pp. 251–256. IEEE (2015)

88

C. Gerdenitsch et al.

22. Kim, L.H., Follmer, S.: Ubiswarm: Ubiquitous robotic interfaces and investigation of abstract motion as a display. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 1(3) (2017). https://doi.org/10.1145/3130931, https://doi.org/10.1145/ 3130931 23. Kipnis, D., Schmidt, S.M., Wilkinson, I.: Intra-organizational influence tactics: explorations in getting one’s way. J. Appl. Psychol. 65(4), 440–452 1980. https:// doi.org/10.1037/0021-9010.65.4.440 24. Lee, S., Han, S., Cheong, M., Kim, S.L., Yun, S.: How do i get my way? a metaanalytic review of research on influence tactics. Leadersh. Q. 28(1), 210–228 (2017). https://doi.org/10.1016/j.leaqua.2016.11.001 25. Lowe, K.B., Kroeck, K.G., Sivasubramaniam, N.: Effectiveness correlates of transformation and transactional leadership: a meta-analytic review of the mlq literature. Leadersh. Q. 7, 385–425 (1996). https://doi.org/10.1016/S10489843(96)90027-2 26. Lunghi, G., Marin, R., Di Castro, M., Masi, A., Sanz, P.J.: Multimodal humanrobot interface for accessible remote robotic interventions in hazardous environments. IEEE Access 7, 127290–127319 (2019). https://doi.org/10.1109/ACCESS. 2019.2939493 27. Ma, L.M., IJtsma, M., Feigh, K.M., Pritchett, A.R.: Metrics for human-robot team design: A teamwork perspective on evaluation of human-robot teams. J. Hum.Robot Interact. (2022). https://doi.org/10.1145/3522581 28. Madigan, C., Johnstone, K., Way, K.A., Capra, M.: How do safety professionals’ influence managers within organizations? a critical incident approach. Saf. Sci. 144, 105478 (2021) 29. Marin Vargas, A., Cominelli, L., Dell’Orletta, F., Scilingo, E.P.: Verbal communication in robotics: A study on salient terms, research fields and trends in the last decades based on a computational linguistic analysis. Front. Comput. Sci. 2 591164 (2021). https://doi.org/10.3389/fcomp.2020.591164 30. Saunderson, S., Nejat, G.: How Robots Influence Humans: a survey of nonverbal communication in social human–robot interaction. Int. J. Soc. Robot. 11(4), 575– 608 (2019). https://doi.org/10.1007/s12369-019-00523-0 31. Saunderson, S.P., Nejat, G.: Persuasive robots should avoid authority: The effects of formal and real authority on persuasion in human-robot interaction. Science robotics 6, eabd5186 (2021). https://doi.org/10.1126/scirobotics.abd5186 32. Schriesheim, C.A., Hinkin, T.R.: Influence tactics used by subordinates: a theoretical and empirical analysis and refinement of the kipnis, schmidt, and wilkinson subscales. J. Appl. Psychol. 75, 246–257 (1990) 33. Thomas, F., Johnston, O.: The illusion of life: Disney animation. Hyperion New York (1995)

OptimizingMARL: Developing Cooperative Game Environments Based on Multi-agent Reinforcement Learning Tha´ıs Ferreira1(B) , Esteban Clua1 , Troy Costa Kohwalter1 , and Rodrigo Santos2

2

1 Universidade Federal Fluminense, Niteroi, Brazil thais [email protected], {esteban,troy}@ic.uff.br Universidade Federal do Estado do Rio de Janeiro, Rio de Janeiro, Brazil [email protected]

Abstract. Intelligent agents are critical components of the current game development state of the art. With advances in hardware, many games can simulate cities and ecosystems full of agents. These environments are known as multi-agent environments. In this domain, reinforcement learning has been explored to develop artificial agents in games. In reinforcement learning, the agent must discover which actions lead to greater rewards by experimenting with these actions and defining a search by trial and error. Specifying when to reward agents is not a simple task and requires knowledge about the environment and the problem to be solved. Furthermore, defining the elements of multi-agent reinforcement learning required for the learning environment can be challenging for developers who are not domain experts. This paper proposes a framework for developing multi-agent cooperative game environments to facilitate the process and improve agent performance during reinforcement learning. The framework consists of steps for modeling the learning environment and designing rewards and knowledge distribution, trying to achieve the best environment configuration for training. The framework was applied to the development of three multi-agent environments, and tests were conducted to analyze the techniques used in reward design. The results show that the use of frequent rewards favors the emergence of essential behaviors (necessary for the resolution of tasks), improving the learning of agents. Although the knowledge distribution can reduce task complexity, dependency between groups is a decisive factor in its implementation.

Keywords: Multi-agent reinforcement learning environments · Games

· Cooperative

This work is supported by CAPES and FAPERJ. c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 89–102, 2022. https://doi.org/10.1007/978-3-031-20212-4_7

90

1

T. Ferreira et al.

Introduction

The use of Artificial Intelligence (AI) techniques in electronic games has been notably explored in recent years [15,19]. This trend can be explained by the constant advances in research, the evolution in hardware and processing, and the consumer market demands. Reinforcement Learning (RL) [10] is an AI technique that has been largely applied and tested in games [20]. In RL, agents learn through rewards. If the agent receives a positive reward, it begins to understand that this behavior is beneficial. Similarly, if it receives a negative reward, it understands that something is wrong and looks to avoid this behavior. Through these rewards, the agent can learn, so the reward design must be carried out effectively, especially in multi-agent environments, where the distribution of rewards should favor collaboration among agents. Understanding the whole process of multi-agent reinforcement learning is not a simple task. Although many works explore the Multi-Agent Reinforcement Learning (MARL) domain [2,4,5,9,16,18,19,21], the focus is not on game development but rather on creating new environments or algorithms. These environments are scaled for testing in specific domains with a focus on the performance of the algorithms used [1,8,11,12]. Exploring the MARL domain in game development is extremely important and can bring new perspectives and approaches. Moreover, designing rewards for the best agent learning is not an easy task, and it is necessary to explore and present methods around this challenge. In this context, we propose the OptimizingMARL, a framework for cooperative game environments development based on MARL. The framework brings practices that guide the development and configuration of cooperative multiagent environments looking for the best performance in agent training. This performance is related to the use of dense rewards (frequent rewards) to encourage the emergence of essential behaviors - those necessary to solve the tasks and the problem. It is possible to identify how the reward design influences the agents’ training performance and which strategies are best for each environment developed through the experiments performed. Furthermore, the framework consists of steps for modeling the learning environment, enabling macro understanding of RL and game elements. This paper is organized as follows: Sect. 2 outlines related work. Section 3 presents the OptimizingMARL. Section 4 describes the conducted experiments, presenting the development of the environments applying the framework and the analysis of the results. Finally, Sect. 5 concludes this work, listing contributions, limitations, and future work.

2

Related Work

In the context of MARL, many works explore the use of new algorithms to improve agent learning performance [1,8,11,12]. Foerster et al. [5] present the COMA algorithm that uses the same centralized network for all agents, with the agents’ shared observations and actions as input. COMA uses the centralized

OptimizingMARL

91

network to estimate how much each agent influenced the expected value. Each agent learns via a modeled reward that compares the overall reward to the received when that agent’s action is replaced by a default action [5]. Cohen et al. [3] introduce MA-POCA (Multi-Agent Posthumous Credit Assignment), a new algorithm that is very similar to COMA, but is capable of training groups of agents to solve tasks in which individual agents can be removed or introduced during the episode. Iqbal and Sha [7] present the MAAC algorithm that trains decentralized policies in multi-agent settings using a centralized network capable of sharing and selecting relevant information for each agent at every instant. While these works test the algorithms in environments that simulate games, the focus is not on game development. This works does not specify how the reward design was performed, as the focus is on the results and performance of each algorithm. Through the literature review, we found only one study by Zhao et al. [22] that most closely resembles our proposal. Zhao et al. [22] present a hierarchical approach for training agents by breaking complex problems into a hierarchy of easy learning tasks to achieve near-human behavior and a high level of skill in team sports games. The authors first train the agents against other easy-level agents and then make the level more difficult once the agent has learned the basics. For the training, they use a proprietary tool they call Simple Team Sports Simulator (STS2). It simulates a 2D soccer environment, in which the state space consists of the coordinates of the players, their speeds, and an indicator of who has the ball. The authors run tests with agents in 1 vs 1 and 2 vs 2 matches using the PPO [14], DQN [13], and Rainbow [6] algorithms. Through the analysis of the results, the authors concluded that sparse score-based rewards are not sufficient to train the agents even at a high level, which requires the application of a more refined reward design. Following the proposal of Zhao et al. [22], we seek to improve agent performance in MARL. However, our focus is on modeling the elements of MARL and on reward design. We propose an easy and more practical way to create multiagent environments that enables a macro understanding of RL elements in the context of a cooperative multi-agent game environment.

3

The Framework OptimizingMARL

Training intelligent agents in cooperative game environments may not be a simple task. The process requires a series of steps and procedures such as specifying the agents, designing rewards, and developing the environment. Moreover, designing rewards that enable faster learning is not trivial, making the process of creating a multi-agent environment optimized for training difficult. To simplify and optimize the process of building and training intelligent agents in cooperative game environments, we propose the OptimizingMARL, which comprises three macro-steps: (1) conceptual modeling of the learning environment;

92

T. Ferreira et al.

(2) reward design and knowledge distribution; and (3) the creation of the environment. Figure 1 illustrates the macro-steps of OptimizingMARL and the next subsections explain them in detail.

Fig. 1. Overview of the macro steps of the OptimizingMARL.

3.1

Model the Multi-agent Learning Environment

In this step, we conceptually specify game and RL elements (e.g., agents, environment). The concept of the game/environment provides critical information about the goals and tasks that the agents must accomplish and about the objects that will be present in the environment (e.g., objects that the agents can interact with, obstacles, enemies). The environment consists of the place where agents will interact and make decisions. The environment must have the necessary characteristics and objects for the agents to interact and solve their tasks. We can create a document specifying all the objects (and their properties) that must be present in the environment. We define the scenario objects - those that are part of the game, such as items, enemies, and obstacles, that the agents will interact with - and the objects’ characteristics - each object of the scenario has its properties such as colliders, texture, physics, among others. The agent is the actor that observes and performs actions in the environment. An agent must have a observation space - the agent collects its observations about the state of the world before making decisions - and actions. Finally, there is the reward signal. The reward signal should not be given all the time, but only when the agent performs an action that is good or bad for solving the problem. The reward signal is how the goals are communicated to the agent, so it needs to be configured so that maximizing the reward generates the desired optimal behavior. The next subsection presents the OptimizingMARL step related to reward design and the possibility of knowledge distribution based on this design.

OptimizingMARL

3.2

93

Design Rewards and Distributing Knowledge

The reward is primal in RL and it works as an incentive mechanism, telling the agent what is correct or not. The RL agent’s goal is to maximize the total reward received during an execution episode. The reward is immediate and defines the characteristics of the problem to be faced by the agent [17]. To know at what times to reward agents, it is essential that developers have mastery of the problem to be solved in the environment. It is common to have the necessary knowledge about the environment in games. However, depending on the complexity of the environment and the tasks, the projection of rewards may not be trivial or easy to visualize. Therefore, the OptimizingMARL comprises a set of activities for understanding the problem to be solved in the environment and for designing rewards. The steps consist of understanding the agents’ goals as well as the tasks required to solve the problem. Figure 2 presents the flow of activities.

Fig. 2. Activity flow of the reward project/knowledge distribution of the OptimizingMARL.

Simple Reward Design. It starts by identifying the primary goals (activity A) through the conceptual modeling of the environment. The primary goals are directly related to the main problem that the group of agents needs to solve. After their identification it specifies a reward for each of these primary goals

94

T. Ferreira et al.

(activity B). The reward design is complete, and the environment can begin to be modeled if the group gains rewards so that they can learn the behaviors essential to solving the problem. However, if the goals are still complex and the rewards are still sparse, this can hinder the agents’ learning, causing training to take longer. Therefore, it is necessary to design dense rewards so that the group gains rewards more frequently to encourage the emergence of essential behaviors. We define these behaviors as those indispensable to solving the task. Dense Reward Design. To define dense rewards, it is first necessary to identify the sub-goals derived from the primary goals (activity C). A reward is specified for each of these sub-goals (activity D). Therefore, the group gains rewards more frequently, learning essential behaviors more immediately. If the agents still need to do many tasks to achieve these sub-goals, it is interesting to identify these tasks (activity E) and reward the group for each task performed (activity F). Rewards for primary goals, sub-goals, and tasks should follow a value hierarchy. The group should be rewarded discretely for completing tasks while completing sub-goals will be rewarded more heavily until they complete the primary goal (higher reward value). Therefore, it prevents agents from getting “stuck” just completing tasks instead of progressing and solving the main problem of the environment. After step D (or F), if the knowledge distribution is not suitable for the situation, we end the process and generate a conceptual environment with dense rewards. Otherwise, we proceed to activity (G) and specify a group for each sub-goal following the dense reward design. Knowledge Distribution. It is necessary to analyze whether the distribution of knowledge will favor training in the environment after specifying the rewards for sub-goals (and optionally the tasks). This step consists in creating a group for each sub-goal. Since each group has a specific goal, the agents may have fewer insights into the environment. After creating a group for each sub-goal, we generate the distributed conceptual environment. 3.3

Create the Learning Environment

The environment development step is the last step of the process. The development consists of creating the necessary elements in the scenario, the agents, programming their behaviors, indicating what to do when an agent is restarted, specifying its sensors, configuring the algorithms and the neural network parameters, and providing the rewards when necessary, among other activities. Unlike the previous two stages, the environment creation step is more specific, as it depends on the platform being used for agent development and training.

4

Experiments

The experiments were developed using the Unity ML-Agents Toolkit [20]. The flexibility of Unity enables the creation of tasks to complex 3D strategy games,

OptimizingMARL

95

physics-based puzzles, or multi-agent competitive games. Unlike other research platforms, Unity is not restricted to any specific gameplay genre or simulation, making it a general platform. We use the MA-POCA [3], a new algorithm designed to train groups of agents to solve tasks. We performed tests for three configurations of the same environment: simple environment, based on a simple reward design (1), dense environment, which has a dense reward design (2), and a distributed environment, which has a knowledge division among different groups (3). In all experiments the agents perceive the environment through ray cast sensors. These sensors can perceive an element through a tag every time a ray intersects an object. If the tag of that object is specified in the sensor, the agent can perceive this object. ML-Agents Toolkit provides a fully connected neural network model to learn from those observations. Figure 3 shows the three environments developed and used in the tests. The first environment (left) was not created entirely by us. This environment comes with Unity-ML agents, but we made changes to the environment and followed the steps in the framework to get the three possible configurations. The following subsections describes each environment in more details and our respective findings.

Fig. 3. Dungeon Escape environment (left); Color Balls (center); Wild World (right).

4.1

Dungeon Escape Environment

In the Dungeon Escape environment, the agents cooperate to escape from a dungeon before the dragon escapes through the portal. The agents need to defeat the dragon, get a key that was dropped, and open the door. The agents must collect the sword and hit the dragon to defeat the dragon. From this concept of the Dungeon Escape environment, we use the framework to model the learning environment (step 1). It defines the scenario objects (e.g., the agents, the dragon, the door, the key, the sword) and their properties (e.g., colliders, texture). After that, we applied step 2 of the OptimizingMARL following the activities to obtain the three configurations for the environment (simple, dense, and distributed). Figure 4 shows the designed reward and distributed knowledge activities. Finally, we create the learning environment (step 3). In step 2, we first identify the primary goal, which is to escape from the dungeon. Then we reward the group when they achieve this goal. Since we applied

96

T. Ferreira et al.

Fig. 4. Dungeon Escape environment rewards and knowledge distribution project.

only the steps related to the simple reward design, this environment is the simple reward environment (1). We continue following the activities to develop the dense reward environment (2). We identify the sub-goal needed to achieve the primary goal: defeat the dragon. Now the group gets rewarded when they defeat the dragon and when they escape from the dungeon. Since the environment is relatively simple, we decided not to divide the sub-goal into tasks and reward the group for them. But this is at the discretion of the development team. We continue following the activities to develop the environment (3). Once we have identified the sub-goals, we can create an agent type responsible for each of these sub-goals. So we specify an agent type responsible for picking up the sword and defeating the enemy, and another agent type responsible for collecting the key and opening the door. So we will have two groups in this environment. In both simple (1) and dense (2) environments, there is one group composed of six agents. All agents have the same behavior. The group gains a reward only when an agent escapes from the dungeon. In the distributed environment (3), the agents are divided into two groups, each composed of three agents. The sword group is responsible for collecting the sword and defeating the dragon. This group gain a positively reward only when it defeats the dragon. The key group is responsible for collecting the key that the dragon drops and for opening the door. This group gain a positively rewarded only when it effectively opens the door. The sword group does not perceive the key, the door, or who has the key. The key group does not perceive the sword, nor who has the sword. Figure 5 presents the results for all environments: (1) simple reward design, (2) dense reward design, and (3) knowledge distribution. The maximum average reward that can be obtained by a group in all environments is 1.0. For analysis

OptimizingMARL

97

purposes, we perform knowledge distribution even though it is not suitable for this environment due to one group’s dependence on the other.

Fig. 5. Results for the three environments in Dungeon Escape experiment.

The cumulative reward group is the mean cumulative episode reward overall agents and should increase during a successful training session. Observing the cumulative reward graph, it is evident that the use of dense rewards (2, green line) improves learning performance if compared to the simple environment (1, red line). In the distributed environment (3), the sword group (light blue line) obtained high rewards. From the graph, the dependency between the groups is clear. The sword group was able to get more rewards because it depends only on itself. The key group (pink line) got smaller rewards because to complete their goal they need the sword group to complete theirs first. The key group gets more rewards as the sword group learns. Entropy is related to how random the decisions of the model are. It should slowly decrease in a successful training session. It occurred for all environments, meaning that the models make less random decisions during the sessions training. The extrinsic value estimate is the mean value approximation for all states visited by the agent. Once these values have converged to the optimal state values, then the optimal policy can be achieved, which should increase during a successful training session. The sword group in environment 3 presented higher values when compared to the others environments. In this way, this group can obtain the optimal policy in fewer steps. Through the analysis, we can conclude that dense sub-goal reward design is the best option for the Dungeon Escape environment. Although the sword group in environment 3 shows better learning, the key group was disadvantaged by the dependency between the groups. In addition, specifying the dense rewards for this environment takes less time than specifying two different agent types for each group. 4.2

Color Balls Environment

In the Color Balls environment, the agents need to get all fifteen balls to the goal (green area). To catch a ball of a specific color, the agent must wear the same

98

T. Ferreira et al.

color uniform. To put on the uniform, the agent needs to get on one of the colored squares. To change the color of the uniform, only climb the square of the desired color. When the agent owns a ball, it cannot change its uniform. We applied the framework to develop the three configurations of the same environment. In the simple (1) and dense (2) environments, there is one group with ten agents. In environment 1, the group is positively rewarded when it completes the primary goal, which is when all fifteen balls are placed into the green area. In environments 2 and 3, the group is positively rewarded for each ball placed in the green area. In environment 3, the agents do not need to correlate the ball color with the uniform color they are wearing because each group has a specific color. The agents only need to understand that they need to be wearing their uniforms to collect the ball. Figure 6 shows the results for the three environments. For better visualization, the graph shows the results of red group in environment 3. All groups in this environment have the same behavior, where the only difference is that each one perceives the ball and the square in the color of its group. For comparison, the maximum average reward that can be obtained by the groups is 5.0 in all environments. As expected, the simple environment (orange line) showed low group reward value because agents are failing to complete the tasks before time runs out. The entropy showed a slight decay but less when it is compared to the other environments. Since agents fail to accumulate rewards, the model continues to work on bad policies, and the estimated value for all states does not increase.

Fig. 6. Results for the three environments in color balls experiment.

This environment is challenging because agents need to understand that they can only pick up a ball if they are wearing the same color uniform. In this case, the vector of observations has an integer variable that corresponds to the color of the ball (1 = yellow, 2 = red, 3 = white, 4 = purple, and 5 = blue). Thus, it is necessary to associate the variable value with the tag in the raycast sensor, making learning more difficult. The group’s performance improved greatly with the design of the dense rewards (blue line). In environment 2, the group is rewarded for each ball placed in the green area. More immediate reward facilitates the association between the uniform color (integer variable) and the ball tag (string in the raycast sensor).

OptimizingMARL

99

In environment 3, we changed the integer variable to boolean, indicating whether the agent wore the uniform or not. In addition, the agents have decreased their perceptions and no longer need to associate the colors with the balls. Each group has a specific color and the agents need to put on their uniform, pick up the balls, and place them in the goal. This environment showed the best results, reaching the maximum reward starting at 480k steps. For the value estimate, the groups in environment 3 presented higher values when compared to environments 1 and 2. As such, the groups can obtain the optimal policy in fewer steps. The analysis allowed us to conclude that distributing knowledge is the best option for the Color Balls environment. The groups in this environment have the same perceptions and behavior, making the knowledge distribution a simple task. The main change is in the vector of observations that starts working with a boolean variable instead of an integer. 4.3

Wild World Environment

In this environment, the agents of a village must survive for an “indeterminate” amount of time in a hostile environment. It means that while the agents keep reaching the goals, they can accumulate rewards until the maximum number of steps in the episode is reached. These goals consist of: making food to feed all the villagers, keeping the fireplaces lit to fight the cold and prevent the villagers from getting hypothermia, and fighting the ferocious wolves that reside in the forest near the village and can attack the villagers. We applied the OptimizingMARL to develop the three configurations of the environment (simple, dense, and distributed). In the simple (1) and dense (2) environments, there is one group consisting of fifteen agents. In environment 1, the group is positively rewarded when it completes the primary goals: prepare food, light fire, and defeat a wolf. In environment 2, the group is positively rewarded when it reaches the primary goals (higher reward) and sub-goals (small reward): plant vegetables, take wood, and equip the sword. In environment 3, the agents are distributed into three groups. The food group is responsible for collecting seed, planting the vegetables, harvesting the vegetables, and preparing the food (higher reward). This group is positively rewarded when it plants vegetables (small reward) and prepares the food. The fire group is responsible for equipping ax, taking wood, and lighting the fire (higher reward). This group is positively rewarded when it takes the wood (small reward) and light the fire. The wolf group is responsible for equipping the sword and defeating a wolf. This group is positively rewarded when it equips the sword (small reward) and defeats a wolf (higher reward). Figure 7 shows the results for the three environments. The wolf group in environment 3 is the only one that has a reward limit. This group gains rewards when an agent equips a sword and when an agent defeats a wolf. However, the number of swords and wolves is finite, making the maximum reward 36 per episode. Regarding the group cumulative reward, the simple environment (dark blue line) presents the lowest values. This can be easily explained because the

100

T. Ferreira et al.

group is rewarded only when it achieves its primary goals. The problem is that for an agent to achieve one of these goals, it needs to perform a series of tasks beforehand.

Fig. 7. Results for the three environments in wild world experiment.

The group in environment 2 shows a huge improvement over the previous group. The sub-goal reward encourages agents to complete the primary goals in stages. The food group and fire group groups in environment 3 were also able to achieve high rewards. However, even with fewer perceptions about the environment, the cumulative reward did not show much improvement compared to the group in environment 2. The wolf group kept the accumulated reward around 15 (36 is the maximum value), showing no great improvement or worsening. The entropy fell for all groups, although it fell more sharply in the food group in environment 3, indicating less random decisions. For the value estimate, the group in environment 2 and the food group and fire group in environment 3 presented higher values when compared to the others. In this way, the groups can obtain the optimal policy in fewer steps. From the analysis, we can conclude that dense sub-goal reward design is the best option for the Wild World environment. The wolf group in environment 3 cannot perform well. Also, specifying the dense rewards for this environment takes less time than specifying three different agent types for each group.

5

Conclusion

In this paper, we introduced a framework to develop multi-agent environments for cooperative games. The proposed framework addresses the design of dense rewards for emergence of essential behaviors and the knowledge distribution. Through the framework it is possible to design environments aiming to improve the agents’ training performance. Based on the analysis of the results, we can conclude that: (i) dense rewards favor the emergence of essential behaviors, improving agent learning; (ii) knowledge distribution can decrease task complexity. The Color Balls environment showed better results with knowledge distribution. It happened because the

OptimizingMARL

101

task became less complex. Instead of associating the value of the integer variable with the ball tag, the agents now have to understand that if they are in uniform (boolean variable is true), they can pick up the ball; (iii) the dependency of the groups is an important factor in the knowledge distribution. In the Dungeon Escape environment, the key group showed lower results compared to the sword group due to their dependency. In the Survival, Color Balls and Wild World environments (food group and fire group), the groups showed very similar results, evidencing the importance of independence; and (iv) the framework allows the development of an environment in order to improve the learning performance of agents in multi-agent environments. The main limitation of this work is the number of environments built and used for testing. At the time this paper was written, we had not yet developed environments that brought more elaborate challenges and incorporated different game mechanics. As future work, we intend to: (i) carry out new tests to analyze the impact of individual rewards (the agent is rewarded and not the group) on the performance of the group as a whole; (ii) build more complex environments; and (iii) seek a way to add communication between groups for environments with distributed knowledge. Acknowledgments. The authors would like to thank NVIDIA, CAPES and FAPERJ for the financial support.

References 1. Bellemare, M.G., Naddaf, Y., Veness, J., Bowling, M.: The arcade learning environment: an evaluation platform for general agents. J. Artif. Intell. Res. 47(1), 253–279 (2013) 2. Berner, C., et al.: Dota 2 with large scale deep reinforcement learning (2019) 3. Cohen, A., et al.: On the use and misuse of absorbing states in multi-agent reinforcement learning (2021) 4. Foerster, J., Assael, Y., Freitas, N., Whiteson, S.: Learning to communicate with deep multi-agent reinforcement learning. In: Proceedings of the 30th International Conference on Neural Information Processing Systems, pp. 2145–2153. NIPS’16, Curran Associates Inc. (2016) 5. Foerster, J., Farquhar, G., Afouras, T., Nardelli, N., Whiteson, S.: Counterfactual multi-agent policy gradients. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 32, pp. 2974–2982. AAAI (2018) 6. Hessel, M., et al.: Rainbow: Combining improvements in deep reinforcement learning. In: Proceedings of the 32nd AAAI Conference on Artificial Intelligence, vol. 32, pp. 3215–3222. PKP Publishing Services Network (2018) 7. Iqbal, S., Sha, F.: Actor-attention-critic for multi-agent reinforcement learning. In: Proceedings of the 36th International Conference on Machine Learning, pp. 2961–2970. PMLR 97, Long Beach, California (2019) 8. Johnson, M., Hofmann, K., Hutton, T., Bignell, D.: The malmo platform for artificial intelligence experimentation. In: Proceedings of the 25th International Joint Conference on Artificial Intelligence, pp. 4246–4247. IJCAI’16, AAAI Press (2016) 9. Jorge, E., K˚ ageb¨ ack, M., Johansson, F., Gustavsson, E.: Learning to play guess who? and inventing a grounded language as a consequence (2016)

102

T. Ferreira et al.

10. Kaelbling, L., Littman, M., Moore, A.: Reinforcement learning: a survey. J. Artif. Intell. Res. 4, 237–285 (1996) 11. Kempka, M., Wydmuch, M., Runc, G., Toczek, J., Ja´skowski, W.: Vizdoom: A doom-based ai research platform for visual reinforcement learning. In: IEEE Conference on Computational Intelligence and Games (CIG), pp. 1–8 (2016) 12. Mnih, V., et al.: Playing atari with deep reinforcement learning (2013) 13. Mnih, V., et al.: Human-level control through deep reinforcement learning. Nature 518, 529–533 (2015) 14. Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O.: Proximal policy optimization algorithms (2017) 15. Summerville, A., et al.: Procedural content generation via machine learning (pcgml). IEEE Trans. Games 10(3), 257–270 (2018) 16. Sunehag, P., et al.: Value-decomposition networks for cooperative multi-agent learning based on team reward. In: Proceedings of the 17th International Conference on Autonomous Agents and MultiAgent Systems. pp. 2085–2087. AAMAS ’18, International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC (2016) 17. Sutton, R., Barto, A.: Reinf. Learn.: Introduction. MIT Press, London, England (2018) 18. Vidhate, D., Kulkarni, P.: Enhanced cooperative multi-agent learning algorithms (ecmla) using reinforcement learning. In: 2016 International Conference on Computing, Analytics and Security Trends (CAST), pp. 556–561. IEEE (2016) 19. Vinyals, O., et al.: Grandmaster level in Starcraft II using multi-agent reinforcement learning. Nature 575, 350–354 (2019) 20. Yannakakis, G., Togelius, J.: Artificial Intelligence and Games. Springer (2018). https://doi.org/10.1007/978-3-319-63519-4 21. Zhang, Q., Zhao, D., Lewis, F.: Model-free reinforcement learning for fully cooperative multi-agent graphical games. In: 2018 International Joint Conference on Neural Networks (IJCNN), pp. 1–6. IEEE (2018) 22. Zhao, Y., Borovikov, I., Rupert, J., Somers, C., Bierami, A.: On multi-agent learning in team sports games. In: Proceedings of the 36th International Conference on Machine Learning (ICML) (2019)

Game Engine Comparative Anatomy Gabriel C. Ullmann1(B) , Cristiano Politowski1 , Yann-Ga¨el Gu´eh´eneuc1 , and Fabio Petrillo2 1 Concordia University, Montreal, QC, Canada g [email protected], c [email protected], [email protected] 2 ´ Ecole de Technologie Sup´erieure, Montreal, QC, Canada [email protected]

Abstract. Video game developers use game engines as a tool to manage complex aspects of game development. While engines play a big role in the success of games, to the best of our knowledge, they are often developed in isolation, in a closed-source manner, without architectural discussions, comparison, and collaboration among projects. In this work in progress, we compare the call graphs of two open-source engines: Godot 3.4.4 and Urho3D 1.8. While static analysis tools could provide us with a general picture without precise call graph paths, the use of a profiler such as Callgrind allows us to also view the call order and frequency. These graphs give us insight into the engines’ designs. We showed that, by using Callgrind, we can obtain a high-level view of an engine’s architecture, which can be used to understand it. In future work, we intend to apply both dynamic and static analysis to other open-source engines to understand architectural patterns and their impact on aspects such as performance and maintenance.

Keywords: Game engines development

1

· Game engine architecture · Game

Introduction

Game engines allow developers to create games in an agile and standardized way. While there is a wide array of free, open-source engines, many large studios choose to develop proprietary solutions to be used internally. Famous examples of this approach are Frostbite1 , id Tech2 , and RAGE3 . There are also popular closed-source engines available publicly, such as Unity and Source [11]. When writing a proprietary engine, developers are able to customize it to fulfill performance or feature requirements of a certain game or game genre, 1 2 3

https://www.ea.com/frostbite. https://arstechnica.com/gaming/2018/08/doom-eternal-reveals-new-powers-putshell-back-on-earth/. https://www.inverse.com/gaming/gta-6-leaks-new-rage-engine.

c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 103–111, 2022. https://doi.org/10.1007/978-3-031-20212-4_8

104

G. C. Ullmann et al.

which may give them an edge on competitors. As a downside, however, this development process hinders exchange of information that could be beneficial to the entire game development community. According to [12], “almost all relevant game engines are closed source which means that particularities on implementation of certain functionalities are not available to developers. The only way for a developer to understand the way certain components work and communicate is to create his/her own computer game engine”. Developers may also find obstacles when creating a game engine from scratch, especially due to a lack of standards on how high-level architecture components should be created and related for this kind of system. Even though books have been published on the topic of game engine architecture [4–6], according to [1], often these publications “tend to only briefly describe the high-level architecture before plunging straight down to the lowest level and describing how the individual components of the engine are implemented”. Furthermore, they mention that “such literature offers an excellent source of information for writing an engine, but provides little assistance for designing one when the requirements are different from the solution described”. We believe game engines would benefit from experience exchanges with other projects of the same kind. A comparative analysis would allow us to identify commonalities and propose points of improvement to existing engines. In this work in progress, we compare the call graphs of two open-source engines, Godot and Urho3D, considering these aspects. The Callgrind profiling tool is used to generate these call graphs, which reflect the execution of a base game project produced with each engine. By comparing the call graphs, we aim to observe the main components of each engine and how responsibilities are divided among them. While static analysis tools could provide us with a general picture without precise call graph paths, we chose to use a profiler because it allows us to view the call order, frequency, and the number of CPU cycles taken by each method. As a result of this analysis, we will answer the following questions: Are game engine designs similar? If so, how similar are they? Our hypothesis is that game engines follow a similar design and architectural structure. We show that producing a high-level architecture view of the engine is possible using a profiling tool and that it is a way to get insights into an engine’s design. In future work, we will compare these architectures with those proposed by researchers, both through static and dynamic analysis. We will study how design choices made in each case may influence performance, maintenance, and the range of games that can be produced with a given engine.

2

Related Work

Several works compare game engines concerning ease of use [3], available tools and target platforms, and also to determine which are more suitable for a given platform [9] or game genre [10]. However, these comparisons are made from a game developer point of view and do not encompass details related to engine

Game Engine Comparative Anatomy

105

design or implementation. Other works focus on proposing architecture to fulfill a specific requirement, such as low-energy consumption for mobile devices [2]. Novel distributed architectures have also been proposed [7,8]. In our work, we compare Godot and Urho3D, not simply in terms of what features are available and in what situations they fit best, but also looking at how engine subsystems (e.g., graphics, audio, physics, etc.) are organized, initialized and in which ways they interact and relate.

3

Approach

3.1

Overview

Our approach consists of five steps. First, we selected C++ engines available on Github (more details on Sect. 3.2) and cloned their repositories. Following the documentation or running scripts provided by the developer, we compiled them. Using the engine’s editor, we created a new project, setting up the minimum necessary to make it run. We will hereby call this our “base game”4 . A Godot project demands at least one scene object to run, which can be created via the engine’s editor. Urho3D, on the other hand, has no editor: one must create a .cpp file and include Urho3D’s library to create a new project. We can run the game loop by calling the Start method, no scene creation needed. We then ran the base game’s packaging process, which allows us to obtain an executable that can be analyzed by the profiler. In the case of Urho3D, there is no packaging, only .cpp file compilation. Normally the code would be compiled in release mode, with all optimizations in place, but in our case, we changed the packaging settings to obtain an executable with debug symbols. We ran Callgrind on the base game executable to obtain the list of all calls made by the program, which it saves into a log file. Using KCachegrind, we converted this Callgrind log into a visual representation of the program’s execution, a call graph. Class and method names in the graph guided our analysis since they helped us understand the responsibilities of each component. 3.2

Engine Selection

We searched for all repositories on GitHub filtering by topic (game-engine) and language (C++). We chose to consider only C++ since this is one of the most popular languages for game engine development [11]. After that, we selected 15 projects and ordered them by number of stars and forks. We then proceeded to clone and build each one of them, starting by those with the highest sum of stars and forks. At the time of writing of this paper, we managed to compile and produce a running executable for two of these engines: Godot and Urho3D.

4

https://github.com/gamedev-studies/engine-profiling-projects/tree/master/ BaseGame.

106

3.3

G. C. Ullmann et al.

Engine Compilation and Analysis

We cloned the GitHub repositories for Godot and Urho3D, working with the latest version at the time for each: commit feb0d at the master branch for Urho3D and commit 242c05 at the 3.4.3-stable branch for Godot. Following the instructions in the documentation, we compiled Godot export templates for Linux (we used Ubuntu 20.04.4 LTS), both in release and debug modes. These are binary files containing engine code compiled for a specific target platform. Before generating an executable for any Godot project, we must link these templates to the project via the editor’s GUI. We created a Godot base game, generated the executable, and then ran it with Callgrind for 30 s.

4

Godot’s Call Graph

We verified that the first tasks Godot executes upon running a game are registering classes and initializing a window (Fig. 1). Since our base game is running on Ubuntu, Godot calls the X11 windowing system. After getting a window instance, the engine then proceeds to call methods to create a scene. Once ready, a notification is sent into the engine’s message queue. According to the engine documentation, the MessageQueue class is “a thread-safe buffer to queue set and call methods for other threads“5 .

Fig. 1. Godot’s base game call graph, showing which methods are called. 5

https://docs.godotengine.org/en/stable/development/cpp/custom godot servers. html.

Game Engine Comparative Anatomy

107

Also during startup, Godot’s calls methods to set up its GUI theme system6 , such as initialize theme, even though in our base game it is not being used. Similarly, a ProceduralSky:: generate sky method is called repeatedly even though no sky is drawn. According to Godot’s documentation7 , this class procedurally generates a sky object, which is “stored in a texture and then displayed as a background in the scene”. Since no textures are loaded into the base game, nothing is drawn, but the sky generation method is called and updated anyway. To validate our approach, we compared Godot’s call graph with a “layers of abstraction” diagram posted on Twitter by Godot’s creator Juan Linietsky8 . This comparison can be seen in Fig. 2. Searching by class names, we found matches for all classes, even though the names were not the same (e.g. PhysicsServer and Physics2DServer ).

Fig. 2. Comparison between the call graph for Godot base game and diagram of “layers of abstraction” by Juan Linietsky

The most notorious difference is the class described by Linietsky as Window. There is no class with this name on Godot’s code base. We believe, however, that he referred to windowing systems in general and therefore decided to simply use the term Window since his diagram was platform-agnostic. Also regarding the windowing system, we found no class named DisplayServer on the call graph. To understand what part of our call graph corresponds to this 6 7 8

https://docs.godotengine.org/en/stable/classes/class theme.html. https://docs.godotengine.org/en/stable/classes/class proceduralsky.html. https://twitter.com/reduzio/status/1506266084420337666.

108

G. C. Ullmann et al.

class, we used method names as a reference. Inspecting the source code, we determined that methods get singleton and has feature from the DisplayServer class were called by the ProjectSettings class, which is instantiated by the first setup method. In the base game’s graph, we found this same method name being called by a class named OS.

5

Urho3D’s Call Graph

Differently from Godot, Urho3D initializes its graphics object first and only then calls DSL (Simple DirectMedia Layer) and X11 to open a window, as shown in Fig. 3. However, they both initialize UI-related code even though there is no UI to draw, and register all game object classes even though they are not instantiated. Furthermore, our attention was drawn to the fact that, while Urho3D calls one initialization method per class, Godot calls Main::setup and Main::setup2. These apparently redundant calls are justified by the developer as a way to separate low and high level “singletons and core types”9

Fig. 3. Urho3D’s base game call graph, showing which methods are called.

6

Comparing Godot and Urho3D

In terms of similarities, both Godot and Urho3D have the same features present in the most widely used game engines, such as graphics, audio, and physics. Both engines register all game object classes upon startup and initialize graphics systems, even though there is nothing more than an empty window to draw. 9

Line 342 at https://github.com/godotengine/godot/blob/3.4/main/main.cpp.

Game Engine Comparative Anatomy

109

Fig. 4. Division of responsibility on Godot and Urho3D: initialization methods (orange), class registration (red) and graphics (blue). Call tree ramifications omitted for brevity. (Color figure online)

On the other hand, the engines run graphics and window initialization in different orders. Godot creates the window first, Urho3D does the opposite and uses the DSL library to manage the interaction with X11. Godot has a scene manager and internal messaging system which are initialized even in a simple project such as the base game. Urho3D initializes only graphics and UI in this case.

7

Discussion

We can use call graphs to visualize the division of responsibilities inside the engines during initialization. They also provide an overview of the most important subsystems and at which moment they are registered and called. However, these observations do not give us deep insights into architectural patterns and design choices for specific subsystems that could be useful for game engine developers. To obtain this knowledge, we will generate dynamic and static call graphs, as well as C++ #include graphs for each subsystem in the future. While comparing the engines, we also identified project and runtime environment conditions that could make the profiler generate the call graph differently. These are important to note since they may change our overall view of the architecture. Operating System: There is a possibility that if we ran the profiler on a Windows or Android version of the same base games, we would obtain a different call graph. Each platform has its own system calls, GUI and data types, so the engines account for this by adding platform-specific code. If the call graph would be largely different, this difference could indicate that code evolution and maintainability are compromised.

110

G. C. Ullmann et al.

Active Features: Important class or method names may not be present in the graphs because they have not been called during the analyzed run. The base game only calls methods from the engine “core” and not others referring to more specific features (e.g., networking methods, which would be needed in a multiplayer game). Other Means of Analysis: Results could vary when running our base game with other profilers such as gprof or gperftools. Furthermore, while dynamic analysis can provide us with a precise call graph, it only represents a given run of the program and is therefore only a partial view.

8

Conclusion

In this work in progress, we showed that producing a high-level architecture view of an engine is possible with the use of a profiler, such as Callgrind, which generates engine call graphs. We compared Godot 3.4.4 and Urho3D 1.8 not only by looking at the engine’s subsystems but also in which order and frequency they are called, which would not be possible by applying static analysis alone. As for the research question “are game engine designs similar?”, we concluded that Godot and Urho3D have a similar feature set and therefore could be used to create the same types of games. Their features are the same present in most widely used game engines, but they divide responsibilities and call them in the code in different ways. Also, we identified that both engines initialize subsystems even though they are not used (e.g. UI). While this may consume resources, it seems to be the right choice since these systems are often used. In future work, we intend to compare different open-source engine call graphs, starting with Unreal Engine 5, to understand what architectural design patterns are most frequently applied to them and how similar they are among engines. Also, we will compare these architectures with those proposed by researchers, using static and dynamic analysis, to understand the impact of design on engine performance, maintenance, and feature richness. Acknowledgements. The authors were partially supported by the NSERC Discovery Grant and Canada Research Chairs programs.

References 1. Anderson, E.F., Engel, S., Comninos, P., McLoughlin, L.: The case for research in game engine architecture. In: Proceedings of the 2008 Conference on Future Play: Research, Play, Share, Future Play 2008, New York, USA, pp. 228–231. Association for Computing Machinery. event-place: Toronto, Ontario, Canada (2008) 2. Christopoulou, E., Xinogalos, S.: Overview and comparative analysis of game engines for desktop and mobile devices. Int. J. Serious Games 4(4), 21–36 (2017) 3. Dickson, P.E., Block, J.E., Echevarria, G.N., Keenan, K.C.: An experience-based comparison of unity and unreal for a stand-alone 3D game development course. In: Proceedings of the 2017 ACM Conference on Innovation and Technology in Computer Science Education, Bologna Italy, pp. 70–75. ACM, June 2017

Game Engine Comparative Anatomy

111

4. Eberly, D.H.: 3D game engine design: a practical approach to real-time computer graphics, 2nd edn. Elsevier/Morgan Kaufmann, Amsterdam, Boston (2007) 5. Gregory, J.: Game engine architecture, third edn . Taylor & Francis, CRC Press, Boca Raton (2018) 6. Lengyel, E.: Foundations of game engine development. Terathon Software LLC, Lincoln, California (2016) 7. Maggiorini, D., Ripamonti, L.A., Zanon, E., Bujari, A., Palazzi, C.E.: SMASH: A distributed game engine architecture. In: 2016 IEEE Symposium on Computers and Communication (ISCC), Messina, Italy, pp. 196–201. IEEE, June 2016 8. Marin, C., Chover, M., Sotoca, J.M.: Prototyping a game engine architecture as a multi-agent system. Z´ apadoˇcesk´ a univerzita. In: Computer Science Research Notes (2019) 9. Pattrasitidecha, A.: Comparison and evaluation of 3D mobile game engines. Master’s thesis, Chalmers University of Technology (2014) 10. Pavkov, S., Frankovic, I., Hoic-Bozic, N.: Comparison of game engines for serious games. In: 2017 40th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO), Opatija, Croatia, pp. 728– 733. IEEE, May 2017 11. Politowski, C., Petrillo, F., Montandon, J.E., Valente, M.T., Gu´eh´eneuc, Y.-G.: Are game engines software frameworks? a three-perspective study. J. Syst. Softw. 171, 110846 (2021) 12. Srˇsen, M., Orehovaˇcki, T.: Developing a game engine in c# programming language. In: 2021 44th International Convention on Information, Communication and Electronic Technology (MIPRO), pp. 1717–1722 (2021)

Design of an Extended Reality Collaboration Architecture for Mixed Immersive and Multi-surface Interaction Thiago Porcino(B) , Seyed Adel Ghaeinian , Juliano Franz , Joseph Malloch , and Derek Reilly Graphics and Experiential Media Lab, Dalhousie University, Halifax, NS, Canada {thiago,adelghaeinian}@dal.ca https://gem.cs.dal.ca/ Abstract. EXtended Reality (XR) is a rapidly developing paradigm for computer entertainment, and is also increasingly used for simulation, training, data analysis, and other non-entertainment purposes, often employing head-worn XR devices like the Microsoft HoloLens. In XR, integration with the physical world should also include integration with commonly used digital devices. This paper proposes an architecture to integrate head-worn displays with touchscreen devices, such as phones, tablets, or large tabletop or wall displays. The architecture emerged through the iterative development of a prototype for collaborative analysis and decision-making for the maritime domain. However, our architecture can flexibly support a range of domains and purposes. XR designers, XR entertainment researchers, and game developers can benefit from our architecture to propose new ways of gaming, considering multiple devices as user interfaces in an immersive collaborative environment. Keywords: Extended reality · Immersive visualization Architecture · Augmented reality · Mixed reality

1

· XR

Introduction

We have seen rapid development and increased public interest in mixed reality technologies in recent years [3] in areas such as training, analytics, and entertainment. In 2021, the augmented reality (AR), virtual reality (VR), and mixed reality (MR) market was valued at 30.7 billion U.S. dollars and it is expected to reach close to 300 billion by 2024 [23]. Head-worn displays (HWDs) are one way to achieve immersive mixed reality or virtual reality. These devices usually consist of electronic displays and lenses that are fixed over the head toward the eyes of the user. HWDs are used in various domains including games and entertainment [24], military [19], education [1], therapy [4], and medicine [12]. While some HWDs are themselves mobile devices (in that they are not tethered to a desktop PC), there is a lack of tools that assist in integrating augmented reality HWDs (such as HoloLens) and other single-user mobile devices c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 112–122, 2022. https://doi.org/10.1007/978-3-031-20212-4_9

Design of an XR Collab. Arch. for Mixed Immersive and MS Interaction

113

(like phones or tablets) with shared devices such as large wall or tabletop displays. In this paper, we present a novel architecture that integrates multiple AR HWDs with a shared large display (a tabletop display in our prototype) and with handheld touchscreen devices such as phones or tablets. AR HWDs are a natural component of multi-system collaborative environments, as they can provide personalized and/or ad hoc extensions to interaction and visualization: for example, a visual representation can be extended beyond or above a tabletop display, gestures and head orientation can become part of a shared interface, and 3D representations of inter-device communication become possible. While the focus of the developed prototype in this work is geospatial data analysis and visualization within a collaborative decision making context, the proposed architecture readily extends to other areas, and in particular entertainment-focused XR applications such as games that integrate with shared displays, such as an interactive mixed reality board game or puzzle. Furthermore, our architecture allows researchers to prototype and study visualizations and interaction techniques that integrate modern HWDs and interactive touchscreens in the general case or in applied contexts such as entertainment, games, training and simulation, immersive analytics, and more. This paper is organized as follows: Sect. 2 describes the related work; Sect. 3 details the iterative design and development of our prototype, leading to the proposed architecture. We then outline future work and current limitations in Sect. 4, and conclude in Sect. 5.

2

Related Work

Multi-system collaborative environments require flexibility of computation, data, interaction, and visualization [9,16]. The general idea is that devices should be dynamically reconfigurable according to a desired or emergent objective. Any multi-system collaborative environment that incorporates AR HWDs should allow for these devices to be flexibly added to, removed from, and configured for the system. Salimian et al. [21] proposed IMRCE, a Unity toolkit for immersive mixed reality collaboration. In this work, users interact with shared 3D virtual objects using touchscreens, in VR, or in mixed reality configurations. IMRCE connects mixed groups of collocated and remote collaborators, and was designed to support rapid prototyping of mixed reality collaborative environments that use hand and position tracking as data. IMRCE was evaluated with groups of developers who developed simple prototypes using the toolkit, and with end users who performed collaborative tasks using it[20]. While IMRCE was compared against a base Unity development environment, other competitive toolkits such as TwinSpace [18], SoD-Toolkit [22], KinectArms [8], or Proximity [14], were not considered, in part due to IMRCE’s embedding within Unity3D. Huh et al. [10] introduce an architecture in which multiple AR/VR clients can collaborate in a shared workspace in a decentralized manner. Their architecture has two data categories: the data stored in the database (based data) and shared

114

T. Porcino et al.

among users, and the user’s data (extension data), which are a modified version of the standard data common to all users. Their architecture facilitates immersive XR collaboration between clients while the network connection is not stable enough using distributed databases and decentralized web technologies. Ran et al. [17] focus on rendering virtual objects around a common virtual point (in the virtual world). They developed a system called SPAR (Spatially Consistent AR). According to the authors, AR apps have issues of high latency, spatial drift, and spatial inconsistency of the virtual assets distributed over time and users. In other words, the virtual objects aren’t rendered simultaneously and with the exact position for all involved users. For this reason, SPAR attends as a new method for communication and computation of AR devices. They worked with an open-source AR platform (Android) instead of closed popular ones such as Apple and Microsoft because their code cannot be changed for tests. Authors mentioned they decreased the total latency by 55% and spatial inconsistency by up to 60% compared to the communication of off-the-shelf AR systems. In summary, the related works take different approaches to integrate systems for collaboration. IMRCE [21], SPAR [17], and Huh’s architecture [10] are the three works closest to our own. However, in IMRCE [21] content is limited to presentation in virtual reality, while in SPAR [17] immersive augmented reality content was not integrated with other devices, such as a tabletop display. Our platform supports both diverse hardware platforms and diverse forms of content and channels of content presentation. Furthermore, Huh et al. [10] designed a decentralized XR architecture quite similar to our architecture concerning AR processing and visualization. However, they do not include shared screens in their architecture. Unlike other works, in our system, collaborators can work around a shared display (SD) or use tablets, HWDs, or screen touchable devices.

3

Materials and Methods

The software framework presented in this paper was originally developed to support collaborative monitoring, analysis, and decision-making involving maritime data. The resulting system combines a shared tabletop display with tablet displays and AR HWDs used by multiple users. A cloud-based repository provides data from multiple sources to our interfaces. These data are used to generate 2D geospatial visualizations presented on the tabletop display, which are augmented by 3D visualizations in augmented reality. Tablet displays permit individual collaborators to query and constrain the data visualization on both their personal HWD and the shared tabletop display. The tablet interface lets individuals query structured data visually, either through touch-based interaction with query primitives on the tablet, interactive selection and filtering of the geospatial visualization on the tabletop, or direct manipulation of 3D objects via the AR HWD. The querying system is built on top of SPARQL [15], a query language for structured databases commonly used by data-driven AI systems.

Design of an XR Collab. Arch. for Mixed Immersive and MS Interaction

115

We used the Unity 3D game engine as the primary development framework to produce the XR content. The first reason was that Unity 3D is a powerful game engine that allows the creation of rich interactive AR/MR/VR/XR experiences (such as games or simulations) for mobile devices. Second, we adopt the idea of exploring the multi-player gaming concept for asset communication between the server and clients in our AR solution. We used a library for multi-player network communication (Photon Engine) and developed the server-side (Shared Server or SS) and a client-side (AR Room). The AR Room can be described as a set of clients that can connect to the server or Shared Server (SS). Moreover, we propose the following architecture (illustrated in Fig. 1) to integrate the SD with multiple HoloLens’ and tablets. The developed applications were constructed using Unity 3D as game engine for 3D models, interaction, and AR communication, which includes Photon Engine for AR multi-player behaviors between SS and clients inside the AR Room. Moreover, we used React [7] and Leaflet [5] for the web-based SD application and SPARQL for database access. We divided our architecture in 5 entities. – Shared Display (SD) - The SD entity represents a shared physical display with the visualization application embedded. The SD has a web-based application developed in React using the Mapbox API [2] to show the Map and AR 3D objects. The SD sends the Map’s latitude/longitude boundaries, zoom, orientation, and rotation information to the SS in real-time (each interaction in the SD is shared among other connected devices). The SD can also display a unique QR code for calibrating the coordinate systems used by the augmented reality layer (Illustrated in the ”AR Room” box in Fig. 1). The SD application can customize the visualization by selecting specific data using a visual query builder to consult the data source through SS. – Shared Server (SS) - the SS is responsible for converting lat/long data received from the database to the relative position in AR context, processing these data, transforming the selection into a structured data file (JSON), and sharing this data with other devices connected to the SS’ network (i.e., Table, Database, and Augmented Reality Content). The SS also receives input queries from the SD (Fig. 1) and any other connected devices and if necessary converts them to SQL. The SS instantiates the corresponding AR 3D object to show in augmented reality on the HoloLens. In each HoloLens we have an Client App that is connected to the SS by a AR Room layer. All AR 3D objects are available to visualized by clients inside the AR Room layer. – AR Room - In this layer, each AR client is synchronized with the SS. Every AR 3D object spawned by SS can be visualized by one or more clients. Client interactions with AR objects can be visualized privately or shared with the entire network. For example, invoking an AR menu is visualized privately on the local client only, whereas spawned AR Objects resulting from the first interaction might be visualized by other connected clients.

116

T. Porcino et al.

Fig. 1. The overall architecture. Data arrives from the right (Database), and is processed by Shared Server (SS). The user can interact with the system via a physical Shared Display (SD) (by physical touching gestures, such as clicking, pinching, pan, zooming) or augmented reality (making gestures on air in a AR context and wearing a HoloLens). AR devices (HoloLens) and new non-immersive devices (such as tablets or smartphones) can connect to the SS and see data in XR or non-immersive visualizations.

– External Device - In this entity, external devices can connect to SS using the same or similar built applications as the SD. While SD is responsible for communicating with SS and sending SD information (such as zoom, orientation, and screen size), the new device can send queries to request data and the results in real time. This allows to users to avoid using the shared table to consult data if they wish, and the 2D application does not need to follow the same structure as the shared visualization on the SD. – Query Translator API - The query API supports and translates between different query representations (currently SQL, SPARQL, and a custom structured visual query language). This intermediate API ensures that our architecture retains the flexibility to work with different software and their particular query languages.

3.1

2D Visualization

The 2D visualization application consists of three parts, the map visualization, the visual query builder, and the text query editor where users can write their valid SPARQL queries on the query editor (see on top-right on Fig. 2) and see

Design of an XR Collab. Arch. for Mixed Immersive and MS Interaction

117

respective query visuals on the visual query builder. Moreover, the result from both queries (visual or text) allows users to visualize selection areas on the map and their respective assets (2D objects linked to the specific latitude and longitude coordinates). The 2D visualization app supports real-time updates, which means the visual query builder, text query editor, and the map have event listeners and triggers to update the visualization during users interactions (e.g., when a user is selecting a region by visual query builder, they see the results on the visualization area instantly). Additionally, to enable AR devices to see holographic visualization aligned with the SS, we put a QR Code in the application to support the alignment by AR Devices. AR Devices can use this QR Code as a world reference to calculate the relative size and position to align their AR content.

Fig. 2. Shared Display (SD) application developed in React. At the bottom left, a QR code is used to align the AR visualization with the SD. At the top right is the visual query builder, where the user can make data selection in real-time and see the results on the SD.

On top of the map layer (illustrated in Fig. 2), we implemented an Interaction Layer (IL). On the IL, the user can interact with the map using touches, drag and drops, zoom, pans, tangible object placements, or AR interactions using AR tools (e.g., HoloLens). We pass direct interaction data between the SD and the 2D Application using TUIO [11] over a socket connection. This protocol is commonly used for multitouch surfaces and supports both touch and interaction using tangible objects. The IL can process information received from AR clients through their specific web-socket connection, and the updates are reflected in the visual query and the map in real-time. More detailed, to synchronize the SD application and AR clients, the 2D Application sends the visual query data, the map boundaries of the visible region in SS

118

T. Porcino et al.

(north-west and south-east coordinates in latitude and longitude), and the map current zoom level to the AR clients through SS. We adopted web sockets with JSON-formatted messages to perform this communication between SD and SS. 3.2

AR Visualization

As outlined above, the AR application consists of two layers: a shared server (SS) and the AR Room. The SS is responsible for communicating with the database and making conversions, for example: – Converting latitude and longitude to the Unity coordinate system. We used Mapbox SDK [13] for converting the geographic coordinate system (GCS) from our data source to Unity’ coordinate system in meters. – Computing coordinate system transformations between tabletop display and the AR visualization. – Sending the information to Client-render layer (AR Room), which means the application responsible for generating the augmented reality in each connected AR client. Second, the AR Room, which is responsible for rendering in each connected AR client (Client App) the real-time data processed to the virtual environment (including all game objects). This strategy was adopted to avoid high latency in the Client App (See on bottom-right on Fig. 1). The main duties of each client App are: – Allows user to interact with objects, all interactions are computed in each client, individually. – Render the content in AR. In that sense, different AR clients can connect to the SS and they do not need to recalculate conversions that were already made by the SS. This enables the AR clients see the same virtual environment in collaborative way. To make this network communication between SS and AR Room, we adopt the Photon Engine SDK [6]. Photon is a base layer for multiplayer games and higher-level network solutions. This plugin for Unity solves issues such as matchmaking and fast network communication using a scalable approach. Basically, Photon enable to create a shared-room where each client APP can connect and get the same information from server. Additionally, both (SS and Client APP) are Unity applications but with distinct duties. In summary, each element in AR visualization are created and processed by SS, and multiple instances of Client-render connected to the SS’ shared AR room can see and interact with the same AR content (Fig. 3). Furthermore, we implemented some interactions in an augmented reality app (Client APP) for the able user in the decision-making process during our solution. The first one is the query selection (at left in Fig. 4), where the user can see an augmented reality arch connector between the world map selection on SD

Design of an XR Collab. Arch. for Mixed Immersive and MS Interaction

119

Fig. 3. Users are interacting with the SD wearing HoloLens. The HoloLens first-person vision of augmented reality interaction over the SD at the left. A third person of the user interacting with both systems is on the right.

visualization in 2D (built in the query builder) and the 2D object that represent this selection on query builder. In the other interaction (Fig. 4, at the right), the user is grabbing an instance of an augmented reality object to show to others or get more information about the specific vessel.

Fig. 4. The user selects a piece of vessel information (the user can grab and drag a the AR object to use this element to get more details about the selected vessel).

4

Future Work and Limitations

While the presented architecture is finished and can fit in distinct contexts, the developed prototype is under development, and there are still many challenges to overcome. Although this prototype is not a final solution, we believe this work can contribute to other researchers giving insights and directing them to develop multiple-system XR experiences for different purposes (analysis, entertainment, simulations) and fields (health, military, educational, games). While our study case is related to a non-entertainment context, we designed our architecture to support numerous application contexts for analysis, simulation, or entertainment applications (such as XR gaming experiences). In

120

T. Porcino et al.

other words, our proposed architecture is flexible in terms of application context and field. Besides, we are still limited in terms of developed interactions. Currently, we are working to produce more interactions among systems. While the already developed interactions can be a good study case, we realized we need to design more AR interactions before starting the usability tests phase with users. For this reason, the next step of this project is to conduct a user-experimental set of tests and a profound study about user behaviors in our XR interaction interfaces. We also intend to include design thinking techniques to help us construct a memorable experience for the final user of this project. Moreover, it is necessary to conduct an in-depth evaluation of users’ behaviors using our solution and evaluate how this solution contributes to the decision-making process in naval organizations.

5

Conclusion

We presented an architecture to integrate augmented reality with physical SD in this work. We implemented this solution to facilitate the decision-making processes in naval organizations under monitoring vessels’ role. Moreover, we develop SD and AR device systems that allow multiple users to collaborate using immersive devices (e.g., HoloLens) or non-immersive such as tablets or smartphones. Besides, we designed and implemented different ways to visualize data (e.g., visual query builder, touchable gestures, AR gestures). In other words, our novel architecture enables multiple users to see the request and see data in an immersive or non-immersive way using particular devices (AR devices, tablets, and shared displays, which include tabletop and other touchable screens). In terms of flexibility, our architecture allows users to add or remove the immersive layer without affecting the visualization for other users. While the SD is the main non-immersive layer and is dependent on visualization, external non-immersive devices are not dependent and can be connected or disconnected at any time without affecting the visualization. Furthermore, our proposed architecture works with a layer (Shared Server or SS) that helps to avoid unnecessary computational processing in HoloLens (concerning the limited hardware’ memory and graphical processing). Moreover, the interaction of multiple users using immersive and non-immersive devices can produce rich discussion among users. We believe our architecture and prototype can help XR designers and researchers to propose new visualizations in immersive environments that combine multiple devices to facilitate decision-making processes for different purposes (simulation, education, gaming, or analysis).

Design of an XR Collab. Arch. for Mixed Immersive and MS Interaction

121

References 1. Ahir, K., Govani, K., Gajera, R., Shah, M.: Application on virtual reality for enhanced education learning, military training and sports. Augmented Hum. Res. 5(1), 7 (2020) 2. Cadenas, C.: Geovisualization: integration and visualization of multiple datasets using mapbox. California Polytechnic State University (2014) ´ Garcia-Fandino, R.: An immersive journey to the molec3. Calvelo, M., Pi˜ neiro, A., ular structure of sars-cov-2: Virtual reality in covid-19. Comput. Struct. Biotech. J. 18, 2621–2628 (2020) 4. Carri´ on, M., Santorum, M., Benavides, J., Aguilar, J., Ortiz, Y.: Developing a virtual reality serious game to recreational therapy using iplus methodology. In: 2019 International Conference on Virtual Reality and Visualization (ICVRV), pp. 133–137. IEEE (2019) 5. Crickard III, P.: Leaflet. js essentials. Packt Publishing Ltd. (2014) 6. Exit Games Inc: Realtime intro | photon engine. https://doc.photonengine.com/ en-us/realtime/current/getting-started/realtime-intro. (Accessed 22 Apr 2022) 7. Gackenheimer, C., Paul, A.: Introduction to React. Apress, Berkeley, CA (2015). https://doi.org/10.1007/978-1-4842-1245-5 8. Genest, A.M., Gutwin, C., Tang, A., Kalyn, M., Ivkovic, Z.: Kinectarms: a toolkit for capturing and displaying arm embodiments in distributed tabletop groupware. In: Proceedings of the 2013 Conference on Computer Supported Cooperative Work, pp. 157–166 (2013) 9. Herz, D., Lee, P., Lutz, L., Stewart, M., Tuell, J., Wiig, J., et al.: Addressing the needs of multi-system youth: Strengthening the connection between child welfare and juvenile justice. In: Center for Juvenile Justice Reform, pp. 1–69 (2012) 10. Huh, S., Muralidharan, S., Ko, H., Yoo, B.: Xr collaboration architecture based on decentralized web. In: The 24th International Conference on 3D Web Technology, Web3D 2019, p. 1–9. , Association for Computing Machinery, New York, NY, USA (2019). https://doi.org/10.1145/3329714.3338137,https://doi.org/ 10.1145/3329714.3338137 11. Kaltenbrunner, M., Bovermann, T., Bencina, R., Costanza, E., et al.: Tuio: A protocol for table-top tangible user interfaces. In: Proceedings of the The 6th Int’l Workshop on Gesture in Human-Computer Interaction and Simulation, pp. 1–5. Citeseer (2005) 12. K¨ uhnapfel, U., Cakmak, H.K., Maaß, H.: Endoscopic surgery training using virtual reality and deformable tissue simulation. Comput. Graph. 24(5), 671–682 (2000) 13. Linwood, J.: Getting started with the Mapbox SDK. In: Build Location Apps on iOS with Swift, pp. 165–178. Apress, Berkeley, CA (2020). https://doi.org/10. 1007/978-1-4842-6083-8 11 14. Marquardt, N., Diaz-Marino, R., Boring, S., Greenberg, S.: The proximity toolkit: prototyping proxemic interactions in ubiquitous computing ecologies. In: Proceedings of the 24th Annual ACM Symposium on User Interface Software and Technology, pp. 315–326 (2011) 15. P´erez, J., Arenas, M., Gutierrez, C.: Semantics and complexity of sparql. ACM Trans. Database Syst. (TODS) 34(3), 1–45 (2009) 16. Powers, J.D., Edwards, J.D., Blackman, K.F., Wegmann, K.M.: Key elements of a successful multi-system collaboration for school-based mental health: In-depth interviews with district and agency administrators. Urban Rev. 45(5), 651–670 (2013)

122

T. Porcino et al.

17. Ran, X., Slocum, C., Tsai, Y.Z., Apicharttrisorn, K., Gorlatova, M., Chen, J.: Multi-user augmented reality with communication efficient and spatially consistent virtual objects. In: Proceedings of the 16th International Conference on emerging Networking EXperiments and Technologies, pp. 386–398 (2020) 18. Reilly, D.F., Rouzati, H., Wu, A., Hwang, J.Y., Brudvik, J., Edwards, W.K.: Twinspace: an infrastructure for cross-reality team spaces. In: Proceedings of the 23nd Annual ACM Symposium on User Interface Software and Technology, pp. 119–128 (2010) 19. Rizzo, A., et al.: Virtual reality goes to war: A brief review of the future of military behavioral healthcare. J. Clin. Psychol. Med. Settings 18(2), 176–187 (2011) 20. Salimian, H., Brooks, S., Reilly, D.: Mp remix: Relaxed wysiwis immersive interfaces for mixed presence collaboration with 3d content. In: Proceedings of the ACM Human-Computer Interaction 3(CSCW), Nov 2019. https://doi.org/10. 1145/3359207 21. Salimian, M., Brooks, S., Reilly, D.: Imrce: A unity toolkit for virtual co-presence. In: Proceedings of the Symposium on Spatial User Interaction, pp. 48–59 (2018) 22. Seyed, T., Azazi, A., Chan, E., Wang, Y., Maurer, F.: Sod-toolkit: A toolkit for interactively prototyping and developing multi-sensor, multi-device environments. In: Proceedings of the 2015 International Conference on Interactive Tabletops & Surfaces, pp. 171–180 (2015) 23. Statista, A.: The statistics portal. Web site (2021). https://www.statista.com/ statistics/591181/global-augmented-virtual-reality-market-size/ 24. Studios, B.G.: The elder scrolls v: Skyrim. Bethesda Game Studios (2015)

Player Behavior and Analysis

Mental Wear and Tear: An Exploratory Study on Mental Fatigue in Video Games Using the Example of League of Legends Ioannis Bikas1 , Johannes Pfau2(B) , Bastian Dänekas1 , and Rainer Malaka1 1

University of Bremen, Bibliothekstraße 1, 28359 Bremen, Germany {bikasio,daenekba,malaka}@uni-bremen.de 2 University of California, 1156 High Street, Santa Cruz, USA [email protected]

Abstract. Playing video games has become a major factor of spending leisure time. As competitive and e-sports games grow in popularity and size, demands of complexity, time investment and cognitive strain to acquire proficiency and keep up with competitors grow with them. Fundamental psychological work stresses the detrimental impact of sinking extensive amounts of time into tasks onto mental health and capabilities to perform in those tasks. However, the effects of prolonged task execution without adequate rest in self-imposed leisure environments (e.g. video games) are largely under-investigated. We therefore seek to investigate these effects as well as their related consequences in the highly competitive scene of e-sports games. Prolonged task execution in tasks that are primarily of a cognitive nature is usually not associated with physical strain or physical fatigue. We therefore primarily investigate effects on cognitive capabilities, like the onset of mental fatigue, as well as mental well-being. Over the course of seven weeks, we monitored the playing routines of League of Legends players and tracked measures of mental fatigue before and after sessions. Together with qualitative remarks, explanations and attitudes towards playing, time investment and skill acquisition, we gathered player reports showing significantly deteriorated moods and dispositions to playing as well as high indications of mental fatigue based on significant occurrences of related symptoms. Additionally, we analyzed the matches played by the participants during the study period and found a gaussian-like distribution of performance relative to the number of successively played matches. We discuss associated problems and propose ideas for methods facilitating more time-efficient alternatives of video game skill acquisition.

Keywords: Mental fatigue

· Video games · E-sports · Game survey

This research is funded by the Klaus Tschira Stiftung. c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. Göbl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 125–139, 2022. https://doi.org/10.1007/978-3-031-20212-4_10

126

1

I. Bikas et al.

Introduction

The video games industry has grown to unprecedented success with competitive games and e-sports covering a considerable share of the revenue1 . Competitive online multiplayer games like League of Legends (3.5 million daily players)2 DotA 2 [34] (640,000 daily players)3 and Counter Strike: Global Offensive [33] (890,000 daily players)4 display unrivaled daily player numbers in particular. These games prominently feature competitive matchmaking modes where players can compete for rank. Reaching a high rank is a desirable goal for many players and often requires a lot of time and dedication to achieve. An intuitive strategy to pursue this is to increase playtime in order to facilitate improvement (in skill as well as rank). It is known in psychology and sports that executing tasks for prolonged periods of time without adequate rest leads to exhaustion and a decrease in performance. In contrast to traditional sports, it seems unlikely that playing video games causes any considerable physical exhaustion. Therefore, interest shifts to effects on cognitive abilities and mental well-being. One particular effect that is mainly related to prolonged task execution is that of mental fatigue [4]. Mental fatigue describes a change in the mental state of a person caused by sustained task performance in cognitive tasks. This change is typically marked by subjective feelings of tiredness, a decline in task performance and a loss of motivation to continue work on the task [19,20,26]. However, whether video games induce mental fatigue is currently under-investigated. As relevant factors concerning performance in video games, mental fatigue has been found to affect attention [3], executive control [20], manual dexterity and anticipation timing [9], decision making and accuracy [32], as well as reaction time [15] (more detailed explanations in Sect. 2). Attention also plays an integral role in skill acquisition as does the ability to distinguish relevant from irrelevant information [17]. Since mental fatigue has been shown to negatively impact both abilities [3,21], it is feasible to assume that it could impair skill acquisition as well. With the above in mind, mental fatigue impairs functions related to performance as well as learning and could thus mean that the intuitive approach of embracing the “grind” (i.e. playing as much as possible) to facilitate skill gain or improve rank is a suboptimal strategy. Moreover, sitting for prolonged periods of time is known to pose health risks [10] and prolonged computer use may contribute to musculoskeletal injury of the neck and upper extremities [7]. Thus, increasing playtime to gain proficiency may not only be ineffective but may also increase risk for the above mentioned health issues. If more effective approaches could be established, this increase in risk may be entirely unnecessary. This paper seeks to investigate whether there is evidence that players of video games experience any of the above mentioned effects. This could establish a basis for further 1 2 3 4

https://newzoo.com/insights/articles/the-games-market-in-2021-the-year-innumbers-e-sports-cloud-gaming/. https://webtribunal.net/blog/league-of-legends-player-count/#gref. https://steamcharts.com/app/570. https://steamcharts.com/app/730.

Mental Wear and Tear

127

exploration of possible effects on player performance and learning (as opposed to perceived effects) and the development of more effective training techniques as well as guidelines for better performance in video games. These developments could likely be extended to comparable digital domains (e.g. surgery using digital interfaces and minimally invasive techniques). Thus, the following research questions are formulated: – Do indications exist that playing demanding video games for prolonged periods of time may induce mental fatigue (or general fatigue)? – Are players aware of possible health risks related to computer use over prolonged periods of time? – Do players attempt to save time by using techniques that make learning more effecient and are they interested in using such techniques? To approach answering these research questions, we tracked twelve League of Legends [29] players over seven weeks using a two-part questionnaire assessing symptoms of mental fatigue and general fatigue. Additionally, a follow-up survey collected 20 responses to achieve a better understanding of the players’ general perception regarding the topic. This exploratory pre-study found significant indications for the onset of both fatigue and mental fatigue after play sessions of League of Legends as perceived by the participants. Participants confirmed increases in playtime to be their main method to gain rank and proficiency and showed an awareness for the existence of specialised training techniques. Physical and mental strain were reported to be common experiences when playing (with some even experiencing gaming-related injuries). Additionally, participants stated a willingness to use efficient training or learning techniques as an alternative to just playing. This work contributes to games user research by recording and observing game-related effects, such as mental fatigue, that may suppress training and skill acquisition, as well as to the areas of (psychological) health and video games, as mentioned effects can arguably impact a player’s well-being and condition

2

Related Work

Mental fatigue describes a change in the mental state of a person caused by sustained task performance in cognitive tasks over prolonged periods of time. This change is typically marked by subjective feelings of tiredness, a decline in task performance and a loss of motivation to continue work on the task [19, 20,26]. Mental fatigue is commonly encountered in everyday life when work on cognitively demanding tasks is required. This is usually resolved by taking a break from the task - an intuitive step since tiredness and loss of motivation occur. Still, mental fatigue has been found to be one of the most frequent causes for accidents in the workplace [2,25]. The detrimental effects on performance associated with mental fatigue are multifaceted in nature. Boksem et al. showed changes in the efficiency of attentional mechanisms that lead to a decrease in the ability to ignore irrelevant information while identifying

128

I. Bikas et al.

relevant information due to mental fatigue [3]. Faber et al. where an evaluation of ERP (event-related potentials) data collected during their experiment confirmed this [21]. It should be noted that stimulus-driven attention was largely unaffected while goal-driven attention decreased significantly. This means that highly automated behavior can be sustained over long periods of time even in spite of the onset of mental fatigue while unexpected situations that require flexible goal-driven responses become less feasible. Linden et al. investigated the effects of mental fatigue on executive control [20]. Executive control can be defined as a basic cognitive process responsible for goal-directed behavior including attention, task switching and functions used to keep goals and goal-related information in mind [1,8,18]. Their studies showed a decrease in the ability to generate and test goal-oriented hypotheses in subjects affected by mental fatigue. Planning ability in general was also found to decrease which was shown by a significant increase in the time needed to execute the first move in the Tower of London tests. The effects of mental fatigue on performance have been observed in multiple different domains. Duncan et al. found that mental fatigue negatively impacts manual dexterity and anticipation timing in trained adults [9]. Negative impacts on decision making and accuracy have been observed in soccer, basketball and table tennis [32]. Basketball players in particular have been shown to suffer from a decline in technical performance when under the effects of mental fatigue [27]. Reaction time has also been found to be affected [15]. Even highly physical endeavours such as endurance sports and swimming showed a decline in performance due to mental fatigue [22,23,28,35]. Lastly, Fortes et al. found that as little as 30 min of playing video games causes mental fatigue and impairs decision-making in amateur boxers [13].

3

Study

7-Week Study: To investigate whether players experience symptoms of mental fatigue (or general fatigue) when playing strenuous games for prolonged periods of time, we asked twelve participants to fill out a two-part questionnaire before and after their play sessions once a week over a period of seven weeks. We chose League of Legends as the target game because it is known to be difficult and popular. The period of seven weeks was chosen to make the results more resistant to both outside influences like general fatigue at a specific day and game-related factors like losing streaks or particularly close games. The study was advertised via email among the students of the affiliated University (redacted for submission) and additionally distributed in gaming-focused Discord communities in order to better reach the target group of regular League of Legends players. Recipients were encouraged to further distribute the information to reach as many players as possible. As a reward for finishing the study, Riot Points vouchers valued at 20 Euro each were advertised in an attempt to decrease the possible number of participants who may drop out because of the length of the study.

Mental Wear and Tear

129

Follow-Up Survey: To gain additional insights into the general perception players have on performance and skill acquisition in video games, a followup survey was conducted using ten questions (see Sect. 3.1). This survey was distributed in online gaming communities via Discord and also used for semistructured interviews with four of the participants of the 7-week study. Recipients were encouraged to further distribute the survey. 3.1

Measures

Two-part Questionnaire used in the 7-Week Study: Due to the lack of an established and evaluated questionnaire that assesses non-pathological mental fatigue for tasks comparable to competitive gaming, the first part of the questionnaire aims to address the most common symptoms of mental fatigue through five questions rated on 5-point Likert scales: 1. 2. 3. 4. 5.

How much aversion do you feel towards continuing playing right now? How would you rate your ability to further concentrate on playing right now? How tired do you feel right now? How would you rate the effort of playing the game right now? How would you rate your motivation to do your best in another match right now?

These questions are directly drawn from the most pronounced symptoms of mental fatigue as described by the related work [4,19,20,26]. The second part features 40 items taken from the shortened Profile of Mood States (POMS) [14] which were also rated on 5-point Likert scales. These items are used to scan for signs of general fatigue and changes in mood. Follow-up Survey: To gain more insight into the players’ general conception regarding skill acquisition in games, performance in games, fatigue in games, common session lengths and willingness to use techniques to gain proficiency more efficiently, semi-structured interviews using ten guiding questions have been conducted. The guiding questions are the following: Q1: What do you think is the most important factor when it comes to getting good at a game? Q2: Did you ever use any kind of technique to get better at a game more quickly/efficiently? (If so, what did you do and did it work?) Q3: If you had the goal to get a high rank in League of Legends as quickly as possible, what would be your approach? Q4: What do you think professional players do differently from regular players in order to become significantly better at the game? Q5: What do you consider a gaming session of average length? Q6: What do you consider a long gaming session? Q7: Given unlimited time to play League of Legends, can you think of any consequences that would have? Q8: How do you think your performance changes during a gaming session?

130

I. Bikas et al.

Q9: Do you sometimes feel fatigued or tired after playing? (Does that happen every time or only under certain circumstances?) Q10: Did you ever experience any gaming-related physical pains or aches? 3.2

Participants

In total, twelve participants took part in the seven-week study. Of these twelve, seven had to be sorted out because they did not produce usable data sets (did not fill out the questionnaires before and after a session but randomly on different days) or because they quit the study preemptively (without submitting any data at all). Of the remaining five participants, four agreed to voluntarily supply demographic data. Information on gender was optional and collected according to best practices with the following options: woman, man, non-binary, prefer not to disclose, prefer to self-describe (open text box). They were split evenly between male and female with an average age of 25. Two of them stated to play every day while the other two stated to play almost every day. Additionally, 20 participants took part in the follow-up survey that was conducted after the seven-week study period had ended.

4

Results

7-Week Study: Results of part one of the questionnaire (five questions aimed at mental fatigue symptoms) are significant for four out of five questions. Using a signed-rank Wilcoxon test for paired values [38], answers given before and after a play session were compared for each question respectively. Participants felt significantly more aversion to continue playing after their session (p < 0.01, W = 39, Cohen’s d = 1.15 [6]), rated their ability to further concentrate on playing significantly lower (p < 0.01, W = 36.5, d = 1.22), felt significantly more tired (p < 0.01, W = 26, d = 0.96) and rated their motivation to do their best in the game significantly lower (p < 0.01, W = 4, d = 1.68) as compared to before they started playing. Participants did not rate the effort of playing the game to be significantly higher. Using a signed-rank Wilcoxon test for paired values, results of part two of the questionnaire (shortened 40 item POMS) were compared using the same methodology. Results were significantly higher for the items in Table 1 after a play session while Table 2 outlines the POMS items that were found to be significantly lower than before. For the remaining POMS items, no significant changes could be found (Tense, Angry, Proud, On Edge, Grouchy, Ashamed, Hopeless, Uneasy, Restless, Annoyed, Discouraged, Resentful, Nervous, Miserable, Bitter, Anxious, Helpless, Satisfied, Bewildered, Furious, Worthless, Forgetful, Vigorous, Uncertain about Things, Bushed and Embarrassed). It should be noted that the participants had a 45% average win rate over the duration of the study. To investigate whether the players’ perception corresponds to an actual decline in performance, the matches played by the participants during the study period were grouped into streaks. Note that we were able to use the data of all 12 participants for this step. We

Mental Wear and Tear

131

Table 1. Shortened POMS items significantly higher after playing. Item

p-Value

W

Cohen’s d

Worn out

p < 0.01

16

d = 1.24

Unhappy

p < 0.01

7

d = 0.9

Confused

p < 0.01

8

d = 0.65

Sad

p < 0.01

6

d = 0.7

Unable to Concentrate

p < 0.01

19

d = 0.93

Fatigued

p < 0.01

21.5

d = 0.87

Exhausted

p < 0.01

34.5

d = 1.07

Weary

p < 0.01

9

d = 0.64

Table 2. Shortened POMS items that were found to be significantly lower after playing.

Item

p-Value

W

Cohen’s d

Lively

p < 0.01

35

d=1

Active

p < 0.01

13

d = 1.24

Competent

p < 0.01

20.5

d = 1.19

Confident

p < 0.01

12

d = 1.2

Full of Pep

p < 0.01

5

d = 0.77

Energetic

p < 0.01

21

d = 1.15

define a match streak as any number of matches that have been played with no more than 20 min break in-between them. 20 min were chosen as a threshold for several reasons: The Riot data base offers game start times (the moment the players enter the match and can move their character) instead of the time a player entered the queue to find another match or the time a player found a new match. Champions (playable Characters) need to be picked and banned before each match which takes up to 10 min5 . Moreover, receiving match rewards and commending teammates for good conduct as well as chatting with the players of both teams and possibly reporting players for bad conduct is another source of delay between matches. In the given data set, streaks of the following length were present: two, three, four, five. With regards to the win rate within those streaks the following has been found: Streaks of length one (241 wins, 294 losses) showed a 45% win rate, streaks of length two (48 wins, 60 losses) showed a 44.4% win rate and streaks of length three (28 wins, 29 losses) showed a 49% win rate. Moreover, streaks of length four (31 wins, 25 losses) showed a 55.4% win rate and streaks of length five (8 wins, 17 losses) showed a decline to 32% win rate. 5

https://leagueoflegends.fandom.com/wiki/Draft_Pick#Champion_Selection.

132

I. Bikas et al.

Follow-up Survey: In total, 20 people completed the follow-up survey. Outcomes for each particular question were classified and analyzed after structured content analysis [24]. This means similar answers to the ten questions (Q1–10) presented in Sect. 3.1 were counted and combined into categories. When asked about the most important factor to become good at a game, 52% of the answers dealt with time investment (practice, time, patience) while 32% of the answers dealt with analysis of gameplay (one’s own gameplay or professional gameplay) in order to learn from mistakes or gain new knowledge. 76% of the participants stated that they do use techniques to get better at games more effectively with 52% of the answers including watching professional gameplay or guides. Three of the participants used mechanical training (last-hitting in League of Legends and an aim trainer for shooters), reporting it to be successful, while two emphasized the importance of analysing replays to identify bad habits. When asked about their approach to gain a high rank in a game as quickly as possible, the most common answers revolved around time investment (playing as much as possible with a 30% share) and efficiency (focusing on a small amount of champions with a 15% share). This is consistent with their thoughts on how professional players become significantly better at games with the most common answers being vast amounts of playtime (41%) and regular analysis of gameplay (22%). Ambition and so called fundamentals (some kind of implicit knowledge or understanding about the workings of a game that can’t be gained easily) were also expressed twice. When asked about the length of their average gaming sessions, answers showed considerable differences spanning from 40 min to six hours (M = 2.34, SD = 1.41). Similarly, perceptions of what a long gaming sessions is ranged from as low as 50 min to up to twelve hours (M = 4.625, SD = 3.361). To our surprise, 40% of the participants only considered sessions of five or more hours as long. Given unlimited time to play a game, the most common consequences named were decline in physical health (30%), boredom (23%) and decline in mental health (20%) while 13% of the answers related to conflicts with other aspects of life (social life, work, university etc.). When asked about changes in their performance during a gaming session, 48% of the answers reported a decline of performance over time. Moreover, 20% of the participants stated to require a warm-up phase to reach peak performance and 16% said their performance is dependent on their mood. Fatigue after playing was experienced by 85% of the participants with 50% of the answers stating long play sessions as the primary reason and 25% referring to losing streaks. When asked about gaming-related pains or aches, 55% of participants stated to not have experienced anything. Among the other participants, hand and wrist pain was the most common problem (55% of the answers) with the rest of the answers being evenly spread between headaches, back pain, eye strain, thumb arthritis and carpal tunnel syndrome. One participant reported to require medical care to get rid of their problem. Lastly, when asked if they would be willing to use exercises or techniques instead of playing if it reduced the time needed to become proficient at a game, 70% of the participants answered affirmatively.

Mental Wear and Tear

5

133

Discussion

The results of the initial questionnaire show significant changes in four of the five most reported symptoms of mental fatigue: significantly increased aversion to continue the task, significantly lower perceived ability to concentrate on the task, significantly increased perceived tiredness and significantly lower willingness to do one’s best. This could indicate the onset of mental fatigue during play sessions of League of Legends and is likely extendable to other demanding and/or competitive video games. The results of the shortened POMS support this by showing significant changes in categories relevant to fatigue and the ability to perform since participants reported to be worn out, fatigued, unable to concentrate, exhausted and weary. Additionally, participants reported to be less lively, less active, less full of pep and less energetic. This could have implications for general player behavior in competitive settings since “grinding” out matches to gain rank could be a lot less effective than it might intuitively appear to be. It could also be relevant for competitive e-sports events and their organisation. The common practice of having the grand finals of a tournament (reaching already multi-million dollar prize pools such as in the case of DotA 2’s International) at the same day as the losers bracket leading up to it might put the team that has to go through multiple matches on the same day before reaching the finals at a significant disadvantage due to fatigue-related decline of performance. Interestingly, participants did not rate the effort of playing the game to be significantly higher after playing which couldn’t support our initial hypothesis. However, a bias for people to perceive actions that align with their self-concordant goals to be easier (even if more effort is invested) could explain this result [37]. There could also be a self-serving bias [12] against stating that playing a game takes effort since this could negatively reflect on how the player’s competence is perceived. Still, even if these observations are not related to mental fatigue as defined by the related work, they still indicate that playing a competitive game like League of Legends for prolonged periods of time may cause fatigue in general. This could be an important insight, not only in the context of games, but other strenuous digital domains that require high performance over long periods of time like surgeries using digitally controlled equipment or security-related activities like flight monitoring. Note that a positive correlation between video game skill and surgical skill has already been observed [31], suggesting that other transfers between these domains may be possible. Lastly, players reported to be significantly more sad and unhappy while feeling less competent and confident. This could be explained well by the participant’s average win rate of 45% during the study period. While it seems intuitive to assume that a representative sample should show a 50% win rate on average, it should be stressed that this is just a snapshot of the players’ performance during the study period. Interestingly, average win rates seem to be below 50% in general for the lower half of the

134

I. Bikas et al.

ranks6 . Since these ranks are populated by around 80% of the player base7 , win rates below 50% appear to be the norm rather than an exception. The results of the follow-up survey provide insight into the participant’s general perception about performance and skill acquisition in League of Legends. According to the participants, the most important factor related to improvement is time investment as it was the most common answer when asked directly, when asked about their approach to gain rank quickly and when asked about the reason why professional players become significantly better. They were nevertheless aware that techniques could be used to make improvement more effective as analysis of gameplay was the second most common answer to Q1 and Q3 while focusing their efforts on a small number of champions to be more time-efficient was the second most common answer to Q2. Interestingly, diminishing returns in terms of skill acquisition when simply adding more playtime or quality of practice were no concerns, since only one person planned to avoid playing too much due to performance decline even though almost half of the participants reported a decline of performance over time (48%) as an answer to Q8. The only answer related to quality was the emphasis of playing in a high-level environment as the third most common answer to explain how professional players become significantly better. This indicates that the participants judge the main difference between a good and an average player to be a quantitative discrepancy in playtime. Yet, 76% of them used techniques to improve in a more specific way instead of using the same time to simply play more. Usage of mechanical training techniques (which were perceived to be successful), map awareness reminders and guides to gain knowledge further indicate that participants think improvements can be made by subdividing general gameplay into a set of specific skills that can in turn be addressed in a specific manner and shows a willingness to use such tools. With average gaming sessions between 1.5 and 3 h, participants showed a willingness to invest a considerable amount of time into the game. The majority of participants reported to experience fatigue after playing, with 50% of them pointing to session length and 25% to losing streaks as the main cause. Additionally, 45% of participants reported experiences with physical pain caused by gaming including hand and wrist pain, injuries, headaches and back pain. It seems that some kind of (perceived) mental and physical strain in certain areas is relatively common among the participants when playing. This is supported by the answers to Q7 (unlimited time to play the game and its supposed consequences) with decline in physical and mental health making up 50% of the answers. This is an interesting result since the participants main approach to gain proficiency is to increase their playtime even though they simultaneously associate increases in playtime with mental and physical strain. More effective and efficient methods of skill acquisition that save time but still produce tangible results could be a solution to this problem. A demand among the participants for such a solution 6 7

https://www.somebits.com//~nelson/lol-stats-notebooks-2016/ League+of+Legends+win+rates+by+tier.html. https://www.esportstales.com/league-of-legends/rank-distribution-percentage-ofplayers-by-tier.

Mental Wear and Tear

135

also exists since 70% answered affirmatively when asked if they would make use of techniques to improve more efficiently instead of just playing. The players’ perception regarding a decline in performance relative to the length of their gaming session was supported by their win rates. While streaks of one, two and three successive matches showed win rates below 50% (45%, 44.4% and 49% respectively), streaks of length four suddenly increased to 55.4%. Interestingly, the increase in win rate between streaks of one, two, three and four matches corresponds to the participants mention of a perceived warm-up phase. Streaks of five successive matches showed a considerable drop in win rate to 32%. This could imply that a point of optimal performance is reached during a play-session after which performance declines. Still, more data needs to be analyzed to gain insight into this aspect. With regards to our research questions the following can be inferred: Playing demanding video games for prolonged periods of time leads to perceived fatigue with significant indications for the onset of mental fatigue. In our sample, win rates decline considerably for sessions of more than three successive matches. Players use techniques like revision, mechanical practice tools, analysis of professional gameplay and guides to make skill acquisition more efficient while being aware of possible health risks associated with increased playtime. Finally, players are willing to use efficient methods instead of playing to gain proficiency.

6

Limitations and Future Work

The results of this work are mainly restricted due to the low number of participants. Even though the results are significant for almost all categories related to mental fatigue (with the exception of no increase in perceived effort) and fatigue in general, it is difficult to see them as anything other than indicative. Furthermore, the data collected describes the players’ perceptions. This does not provide any information on whether these perceptions are rooted in an actual decline of performance. Even though the participants match data showed a clear decline in win rate relative to the number of successively played matches, it is unclear which aspects of performance were compromised and how strongly each individual participant’s performance declined during these matches. A follow-up study with more participants and analysis of match performances on a larger scale is planned to confirm and extend these findings. If individual match performances can be found to decline relative to time played without adequate rest, this could have implications for fairness in e-sports tournaments (as discussed in Sect. 5). Such a detrimental effect on performance would also make the exploration of structured training methods as well as guidelines for optimal performance in this domain particularly important. If effective, these methods could support players in their e-sports endeavours and additionally decrease the risk of gaming-related injuries as well as health and mental health issues by lowering requirements of play time to achieve the same (or possibly better) results. Limitations also apply to the recruitment process. Gaming focused community discord servers were used as the primary venue to advertise the study

136

I. Bikas et al.

in an attempt to preserve ecological validity by reaching actual players from the chosen target group. Albeit difficult to achieve, recruiting players from all the different regions and from all the different ranks (representing skill levels) in equal amounts would facilitate more representative results. Another limitation for this study is the game of choice. Even though the results are likely extendable to other competitive or strenuous games, more genres need to be investigated to really confirm this assumption. Additionally, the multiplayer as well as the team aspect could influence how strongly players fatigue. Studies with single-player games are needed to make the results more relevant for digital domains that require one person to work on a non-competitive task (e.g. flight monitoring or surgery). With regards to future work, it is also unclear whether the same training methods and performance guidelines will necessarily apply to games in other genres. Further improvements are possible regarding the measures used in this study. A fully validated questionnaire for detecting mental fatigue in players of video games specifically would have produced much stronger evidence. To our knowledge, such a questionnaire unfortunately does not exist. Even though this paper argues that mental fatigue might have a negative effect on performance in video games due to its effects on cognitive functions (described in Sect. 2), more empirical evidence is needed to confirm this conjecture. A follow-up study investigating the behavior of performance metrics relative to uninterrupted playtime as mentioned before is thus essential. Since participants stated a demand for techniques and structured methods to improve more efficiently at video games, identification and evaluation of such techniques will be explored in future work. Mental practice and deliberate practice (as defined by Ericsson et al. [11]) are both interesting areas that have been found to be beneficial in other areas like traditional sports or music [5,16,30,36]. Lastly, the digital nature of video games allows the development and use of integrated practice tools for skill acquisition that could make effective practice more seamless and enjoyable. Digital tools could be used to improve feedback, visualize player performance, indicate optimal play patterns, supply gameplay related information on demand or even analyze player performance automatically to detect weaknesses and suggest areas of improvement. Development and evaluation of such tools could not only make skill acquisition in video games more effective but also more accessible, sustainable and fun.

7

Conclusion

Video games are one of the most popular leisure activities in the world. People are willing to spend large amounts of time playing these games, especially in the case of competitive games like League of Legends. Playing long sessions without rest seems to be a popular method to gain rank and proficiency in such games. Since this requires considerable amounts of time, concerns related to the effectiveness of this method are valid. It is known in other fields that executing tasks for prolonged periods of time leads to fatigue and mental fatigue in particular, which can in turn negatively impact performance as well as learning.

Mental Wear and Tear

137

This exploratory study found significant indications for the onset of both fatigue and mental fatigue after play sessions of League of Legends as perceived by the participants together with a considerable decline in win rate when playing five matches in succession. Additionally, participants not only confirmed increases in playtime to be their main method to gain rank and proficiency but also showed an awareness for the existence of specialised training techniques and reported physical and mental strain to be a common experience when playing (with some even experiencing gaming-related injuries). Lastly, participants stated a willingness to use efficient training or learning techniques instead of playing if they enabled them to save time. Acknowledgements. We would like all participants. This research is funded by the Klaus Tschira Stiftung. One author is funded by the James S. McDonnell Foundation (Grant Title: A Methodology for Studying the Dynamics of Resilience of College Students).

References 1. Anderson, J.R., Bellezza, F.S.: Rules of the Mind. Psychology Press, London (2009) 2. Baker, K., Olson, J., Morisseau, D.: Work practices, fatigue, and nuclear power plant safety performance. Hum. Factors 36(2), 244–257 (1994). https://doi.org/ 10.1177/001872089403600206. pMID: 8070790 3. Boksem, M.A., Meijman, T.F., Lorist, M.M.: Effects of mental fatigue on attention: an ERP study. Cogn. Brain Res. 25(1), 107–116 (2005). https://doi.org/10. 1016/j.cogbrainres.2005.04.011, https://www.sciencedirect.com/science/article/ pii/S0926641005001187 4. Boksem, M.A., Tops, M.: Mental fatigue: costs and benefits. Brain Res. Rev. 59(1), 125–139 (2008) 5. Charness, N., Tuffiash, M., Krampe, R., Reingold, E., Vasyukova, E.: The role of deliberate practice in chess expertise. Appl. Cogn. Psychol. 19(2), 151–165 (2005) 6. Cohen, J.: Statistical Power Analysis for the Behavioral Sciences. Routledge, Milton Park (2013) 7. Cook, C., Burgess-Limerick, R., Chang, S.: The prevalence of neck and upper extremity musculoskeletal symptoms in computer mouse users. Int. J. Ind. Ergonomics 26(3), 347–356 (2000). https://doi.org/10.1016/S0169-8141(00)00010X, https://www.sciencedirect.com/science/article/pii/S016981410000010X 8. Duncan, J., Emslie, H., Williams, P., Johnson, R., Freer, C.: Intelligence and the frontal lobe: the organization of goal-directed behavior. Cogn. Psychol. 30(3), 257– 303 (1996). https://doi.org/10.1006/cogp.1996.0008 9. Duncan, M.J., Fowler, N., George, O., Joyce, S., Hankey, J.: Mental fatigue negatively influences manual dexterity and anticipation timing but not repeated highintensity exercise performance in trained adults. Res. Sports Med. 23(1), 1–13 (2015) 10. Dunstan, D.W., Howard, B., Healy, G.N., Owen, N.: Too much sitting-a health hazard. Diabetes Res. Clin. Pract. 97(3), 368–376 (2012) 11. Ericsson, K.A., Krampe, R.T., Tesch-Römer, C.: The role of deliberate practice in the acquisition of expert performance. Psychol. Rev. 100(3), 363 (1993) 12. Forsyth, D.R.: Self-serving bias (2008)

138

I. Bikas et al.

13. Fortes, L.S., et al.: Playing videogames or using social media applications on smartphones causes mental fatigue and impairs decision-making performance in amateur boxers. Appl. Neuropsychol. Adult, 1–12 (2021) 14. Grove, R., Prapavessis, H.: Abbreviated poms questionnaire (40 items), November 2013 15. Habay, J., et al.: Mental fatigue and sport-specific psychomotor performance: a systematic review. Sports Med. 1–22 (2021). https://doi.org/10.1007/s40279-02101429-6 16. Isaac, A.R.: Mental practice: does it work in the field? Sport Psychol. 6(2), 192–198 (1992) 17. Johnson, A., Proctor, R.W.: Skill Acquisition and Training: Achieving Expertise in Simple and Complex Tasks. Routledge, Taylor & Francis Group (2017) 18. Jong, R.: An intention-activation account of residual switch costs. Control Cogn. Process. 18, 356–376 (2000) 19. Linden, D.: The urge to stop: the cognitive and biological nature of acute mental fatigue (2011) 20. Linden, D., Frese, M., Meijman, T.: Mental fatigue and the control of cognitive processes: effects on perseveration and planning. Acta Psychol. 113, 45–65 (2003). https://doi.org/10.1016/S0001-6918(02)00150-6 21. Lorist, M.: How mental fatigue affects visual selective attention. Front. Hum. Neurosci. 5 (2011). https://doi.org/10.3389/conf.fnhum.2011.207.00261 22. Marcora, S.M., Staiano, W., Manning, V.: Mental fatigue impairs physical performance in humans. J. Appl. Physiol. 106(3), 857–864 (2009) 23. Martin, K., Meeusen, R., Thompson, K.G., Keegan, R., Rattray, B.: Mental fatigue impairs endurance performance: a physiological explanation. Sports Med. 48(9), 2041–2051 (2018) 24. Mayring, P., Fenzl, T.: Qualitative Inhaltsanalyse. In: Handbuch Methoden der empirischen Sozialforschung, pp. 633–648. Springer, Wiesbaden (2019). https:// doi.org/10.1007/978-3-658-21308-4_42 25. McCormick, F., Kadzielski, J., Landrigan, C.P., Evans, B., Herndon, J.H., Rubash, H.E.: Surgeon fatigue: a prospective analysis of the incidence, risk, and intervals of predicted fatigue-related impairment in residents. Arch. Surg. 147(5), 430–435 (2012). https://doi.org/10.1001/archsurg.2012.84 26. Meijman, T.F.: Mental fatigue and the efficiency of information processing in relation to work times. Int. J. Ind. Ergon. 20(1), 31–38 (1997). https://doi.org/10. 1016/S0169-8141(96)00029-7, https://www.sciencedirect.com/science/article/pii/ S0169814196000297 27. Moreira, A., Aoki, M.S., Franchini, E., da Silva Machado, D.G., Paludo, A.C., Okano, A.H.: Mental fatigue impairs technical performance and alters neuroendocrine and autonomic responses in elite young basketball players. Physiol. Behav. 196, 112–118 (2018) 28. Penna, E.M., et al.: Mental fatigue impairs physical performance in young swimmers. Pediatric Exerc. Sci. 30(2), 208–215 (2018) 29. Riot Games: League of Legends. Game [PC], riot Games, Los Angeles, California, USA, October 2009 30. Ross, S.L.: The effectiveness of mental practice in improving the performance of college trombonists. J. Res. Music Educ. 33(4), 221–230 (1985) 31. Rosser, J.C., et al.: The impact of video games on training surgeons in the 21st century. Arch. Surg. 142(2), 181–186 (2007)

Mental Wear and Tear

139

32. Sun, H., Soh, K.G., Roslan, S., Wazir, M.R.W.N., Soh, K.L.: Does mental fatigue affect skilled performance in athletes? a systematic review. PLoS ONE 16(10), e0258307 (2021) 33. Valve: Counter-Strike: Global Offensive. Game [PC,PS3,XBox360], valve, Bellevue, Washington State, USA, August 2012 34. Valve: Dota2. Game [PC], valve, Bellevue, Washington State, USA, July 2013 35. Van Cutsem, J., Marcora, S.: The effects of mental fatigue on sport performance: an update. In: Motivation and Self-regulation in Sport and Exercise, pp. 134–148. Routledge (2021) 36. Ward, P., Hodges, N.J., Starkes, J.L., Williams, M.A.: The road to excellence: deliberate practice and the development of expertise. High Ability Stud. 18(2), 119–153 (2007) 37. Werner, K.M., Milyavskaya, M., Foxen-Craft, E., Koestner, R.: Some goals just feel easier: self-concordance leads to goal progress through subjective ease, not effort. Pers. Individ. Differ. 96, 237–242 (2016). https://doi.org/10.1016/j.paid.2016.03. 002, https://www.sciencedirect.com/science/article/pii/S0191886916301477 38. Wilcoxon, F.: Individual comparisons by ranking methods. In: Kotz, S., Johnson, N.L. (eds.) Breakthroughs in statistics, pp. 196–202. Springer, New York (1992). https://doi.org/10.1007/978-1-4612-4380-9_16

Difficulty Pacing Impact on Player Motivation William Rao Fernandes(B)

and Guillaume Levieux

Conservatoire National des Arts et Metiers, CEDRIC, Paris, France {william.fernandes,guillaume.levieux}@cnam.fr

Abstract. Challenge, and thus difficulty, is one of the main factors of enjoyment and motivation in video games. To enhance the players’ motivation, many studies rely on Dynamic Difficulty Adjustment model in order to follow a difficulty curve. However, few authors worked on the shape of the difficulty curve itself. Our goal in this paper is to evaluate how players react to different difficulty curves. We use four different difficulty curves, including two flat curves and two curves with different baseline and peak levels. We test those curves on 67 students of a video games school while playing a First-Person Shooter game. Our study shows that curves with peaks have the strongest impact on players’ motivation.

1

Introduction

Whether when studying video games or psychology, motivation is an important part of the literature. In fact, many studies on video games are using psychology theories in order to enhance the players’ motivation [8,19,20,29]. Many of these studies try to design a difficulty curve that matches the Flow Theory, as the flow state is known as a mental state where players are strongly focused [27]. This is considered as an engaging moment, providing a positive feeling between arousal and control. The flow theory states that difficulty should be neither too hard nor too easy, matching the player skills, and that other interesting states can be obtained by slightly unbalancing difficulty. However, there are no scientific studies on what kind of difficulty curve might enhance the players’ motivation even more. The only articles we found are from game designers who share their point of view about their industrial experiences [11,31,34]. The goal of this paper is to design different difficulty curves in order to enhance the players’ motivation. Then, by using these curves in a video game, compare the players’ actual motivation for each curve. To do so, we will use a Dynamic Difficulty Adjustment (DDA) model which can follow any difficulty curve [28]. The paper starts with a literature review of difficulty and motivation, which help us shape our DDA model. However, few authors have experimentally studied difficulty curves shapes, but there exists theoretical models and design good c IFIP International Federation for Information Processing 2022  Published by Springer Nature Switzerland AG 2022 B. G¨ obl et al. (Eds.): ICEC 2022, LNCS 13477, pp. 140–153, 2022. https://doi.org/10.1007/978-3-031-20212-4_11

Difficulty Pacing Impact on Player Motivation

141

practices that can help us to design our difficulty curves and to form our hypotheses. Then we detail our experiment, present our results and finally discuss them.

2

Difficulty and Motivation

Many authors consider challenge as one of the core features of video games’ enjoyment. Ryan et al. study intrinsic motivation and apply Self-Determination Theory to video games. They show how enjoyment is related to the feeling of competence, which relies on an optimal level of challenge, and thus, on the game’s difficulty [29]. Jesper Juul’s definition of video games states that a video game has quantifiable outcomes, influenced by the players’ effort. This definition puts challenge as part of the very nature of a game, as the level of challenge mainly drives the effort the players have to put in the game. Juul also provided insight on how failure, and thus difficulty, is one of the core aspects of video game enjoyment and learning progression [17,18]. Malone considers that video games are captivating because they provide challenge, foster the player’s curiosity and propose a rich fantasy [25]. Malone explains that challenge is directly related to the game’s difficulty and corresponds to the uncertainty for the players to reach the game’s goals. Lazzaro proposes a four factor model, where Hard Fun is related to the feeling of overcoming difficult tasks [20]. Sweetser et al. see also challenge as one of the most fundamental part of their Game Flow framework [32]. The work of Sweetser et al. stems from Mihaly Csikszentmihalyi’s Theory of Flow [27], who has been trying to figure out the properties of activities showing a strong, intrinsic ability to motivate. Csikszentmihalyi research states that these activities provide perceived challenges, or opportunities for action, that stretch (neither over matching nor underutilizing) existing skills [27]. Such a flow state has been shown to be globally amplified by the use of a basic difficulty adjustment system [3]. A study of large population of players in two commercial games confirm that players prefer specific levels of difficulty [1]. In order to enhance the motivation of the player, many studies focused on Dynamic Difficulty Adjustment (DDA) models. DDA models are used to automatically change the difficulty parameters of a game, based on the players’ skills, in order to keep the players entertained. Many DDA models are using data representing the player’s performance. The +/- δ algorithm, for instance, can start with no data and only use the player’s last success or failure to converge to what we may call a balanced difficulty state, where the player has a 0.5 probability to fail [28]. Constant used this algorithm to adapt the difficulty of three games based on logical, motor and sensory difficulties [8]. This algorithm is also used in the famous game Crash Bandicoot [12]. DDA algorithms can be specific to a game genre, like the Rubber Band artificial intelligence (AI), used for sport or racing games, that will adjust the difficulty by comparing the position of the enemies and the player [37]. We also have more advanced models, using learning algorithms to adapt the difficulty like Andrade with Q-learning [2], dynamic scripting [30], or Monte Carlo Tree Search [9,14,16]. In a previous research, we developed a simple DDA algorithm that needs few data points [28]. The model is

142

W. R. Fernandes and G. Levieux

using both +/- δ algorithm and logistic regression. The +/- δ algorithm is used to gather enough data until the logistic regression is accurate enough, which let the model estimate the parameters needed to provide the targeted failure probability. Difficulty is thus a fundamental aspect of video games, and the progression of difficulty seems to be one of the many ways that video games can keep us motivated and concentrated for long period of times. One way to enhance motivation using difficulty is by designing difficulty curves that represent how difficulty will change over time. By shaping the difficulty curve, designers can decide when to challenge the players, when to give them rest, gradually increasing or decreasing difficulty or creating difficulty peaks. The difficulty curves can also be used by a DDA model, as presented previously. However, there is little to none literature about how to design a good difficulty curve. We thus focus on the difficulty curve’s shape, in order to provide experimental evidence of their impact on player’s motivation.

3

Difficulty Curves

As we said, difficulty is thought over time. Games rarely have the same difficulty level from the beginning to the end [1]. One way to plan the pace of difficulty is to design a difficulty curve that will drive the game’s difficulty during the game session [6]. According to some game designer and also proposed by [21], a good difficulty curve will start low in difficulty, gradually raising the difficulty until a specific event occurs. Then, difficulty can be lowered to let the players enjoy their success, and then gradually increase again to repeat the pattern [11, 31,34]. Often, the difficulty curve will match the introduction of new gameplay mechanics that the player will have to master in order to move forward. That is the case for The Legend of Zelda, an iconic game from Nintendo [26]. The players will get a new item at the beginning of a dungeon, meaning that they will need to learn new skills. This leads to a difficulty peak. Then, as the players explore the dungeon, they will improve their skills, which translate to the difficulty going down. Finally, the players will face the dungeon master, a new peak of difficulty. Getting out of the dungeon, the new item will allow players to reach new parts of the game map, that were not accessible, or unlock shortcuts, and difficulty is thus going down. But these new parts of the game maps may lead to stronger enemies, new dungeons, and thus the difficulty goes up again. Allart et al. worked on the impact of two games difficulty curves on player retention. The data comes from two industrial games, Rayman Legends and Tom Clancy’s: The Division [1]. They showed that difficulty is by itself an explanatory variable of player retention, and that players tend to prefer higher levels of difficulty. It is to note that both games have almost always a probability of failure under the balanced difficulty of 50%. They have also shown that people prefer lower difficulty at the beginning, which confirm the self-efficacy theory [5]. This paper also shows that difficulty peaks are present in Ubisoft games [1]. It is shown that players enjoy difficulty peaks when the game is not punitive

Difficulty Pacing Impact on Player Motivation

143

like Rayman Legends, which comforted us for our experimentation, as difficulty curves and difficulty peaks seemed at least as important as overall difficulty. Loftus studied why gamers are attracted to video games [23]. At this time, players mainly played paying arcade games. Players were rewarded by a high score when playing well, and had to pay when losing in order to keep their progression. The main goal of game designers was to have a difficulty hard enough, so the players lose a lot, but they feel like they can beat the game, so they keep playing. The path to success is reachable but close to failure, and the players might regret their decisions afterward [13]. Regret is a way to make the players try again where they failed, because they were close to the success they wanted. To apply that idea of regret, we can use difficulty curves with some peaks of difficulty. By doing so, we challenge the players to higher difficulty than they are used to, which lead to a greater reward when they clear the level. Weiner described that people react with positive emotions if they feel like their actions are the cause of success [35]. However, Weiner showed that people will feel less positive emotions if the task is too easy, which confirms that the task difficulty is a variable we need to take into account. We can apply that to video games, as the players will feel less satisfied winning a game if they know that the game difficulty was set to easy [24]. Klimmt decided to let the players know in which difficulty they were during his experimentation. However, Klimmt found out that players enjoyed the game more with a lower difficulty, when they had very few failures [19]. It is to note that players had 10 min of gameplay, which leads us to suggest that the first contact with the game should be with a low difficulty. The idea of positive emotion linked to success is also found in a more advance form of the Flow theory, where the authors seek to propose tasks with a difficulty lower than the people skills in order to put people in different positive emotional states [27]. Linehan et al. focused on four mainstream puzzle games, Portal, Portal 2, Braid and Lemmings. They analyzed the data from these games in order to understand why people were attracted to them. For all the games, the same pattern is used. The beginning of the game is really simple, as players need to learn the basic mechanics. Then each time the game includes a new mechanics, they have some time to familiarize with it with a low difficulty puzzle before getting to a more challenging level of difficulty [22]. Following our literature review, we have three major point to design a difficulty curve. The first one is the starting difficulty. Following Linehan et al., as well as Bandura’s self-efficacy theory, each of our curves will start low [5,22]. The second point is the difficulty level. We can follow the flow and suggest a curve that goes up to 50% chance of winning because, as we saw it in the literature review of difficulty and motivation, it is widespread. This will be our balanced curve that we will name Flow Curve as it simply follows the goal of matching the player skills. To test the difference with difficulty level, we designed an 80% chance of winning curve, as players might prefer a lower difficulty [19]. This will be ourLow Curve.

144

W. R. Fernandes and G. Levieux

The third point is the presence or absence of difficulty peaks. We found in the literature review of the difficulty curves that making difficulty peaks is pleasing for the players [11,31,34]. We designed two more difficulty curves, with the same baseline as the first two, but adding high difficulty peaks that reach 80% chances of failure. We thus add the Low Peaks and the Flow Peaks curves. Figure 1 shows the shape of each of the curves we designed for the experiment.

Fig. 1. Designed difficulty curves

The continuous orange curves are flat curves, while the dotted violet ones are curves with difficulty peaks. Difficulty peaks appear at every five rounds of the game, with a failure probability of 0.8.

Difficulty Pacing Impact on Player Motivation

4 4.1

145

Experimentation Hypotheses

Lomas et al. study leads us to suppose that the game difficulty can lead players to a lack of motivation if they know they are playing on low difficulty [24]. We thus chose not to explicitly show the current game’s difficulty. Then we start following the root of the flow theory, with a difficulty that matches the player’s skills. This is however contradictory with Klimmt et al. study, that shows that player might prefer lower difficulty as they provide more positive feedbacks, [19]. The first hypothesis that we will put to test will thus be that the Flow curve, which offers a difficulty corresponding to the player’s skill, might be more appreciated than the Low curve. We are using a game that is not punitive. We will present the game in the Methodology section. In this game, losing leads the players to another try at a similar task, the only difference between the failed task and the new one will be the task difficulty. The players will face the same enemy, in the same arena, with the same items. In this context, following [1] discussion, we believe that the difficulty curves with peaks will be more appreciated, because they will bring more satisfaction if the player succeeds when the difficulty is high, but they will feel very little frustration if they fail, because they can directly start playing again without loss of progress. Finally, our last hypothesis is the grouping of the first two other hypotheses, which is that the Flow Peaks curve will be the most appreciated of the four curves. 1. The Flow curve is more appreciated than the Low curve; 2. The curves with peaks are more appreciated; 3. The Flow Peaks curve is more appreciated than the others. 4.2

Methodology

In this paper and following Levieux and Aponte et al., we consider that an estimation of the game’s difficulty is very close to the estimation of players’ performance, that we define as their failure probability [4,21]. As shown in Fig. 1, the curves represent the failure probability at each try. We modified a Unity First-Person Shooter (FPS) mini-game to make it a one versus one arena game [33]. The Fig. 2 is a screenshot during a play session. In this game, the player can move forward, backward, left and right, and can move the camera using the mouse. We only used one type of weapon, in order to reduce the variability for this experiment. We modified the level design and enemy AI so that the player will fight in an arena versus an AI-controlled enemy. The AI will patrol the map and, when it sees the player, will start to follow them and shoot them. Each time the player beats the enemy, a new enemy respawns, and we refill the player’s health. If the player dies, we destroy the enemy, then respawn the

146

W. R. Fernandes and G. Levieux

player and a new enemy. The AI has different parameters that change according to the difficulty. We can change its movement speed, shooting speed, shooting range, and player detection range. We use a DDA model developed in a previous research to follow our difficulty curve [28]. This DDA model allows us to follow any difficulty curve using as few data points as possible. It can be used on many game types, allows a developer to set the game’s difficulty to any level within approximately two minutes of playtime. In order to roughly estimate the difficulty as quickly as possible, the model drives a single metavariable to adjust the game’s difficulty. As described, the game difficulty depend on the game’s variables which are the enemy moving speed, shooting speed, shooting range and detection range. It starts with a simple +/-δ algorithm to gather a few data points, and then uses logistic regression to estimate the players’ failure probability when the smallest required amount of data has been collected.

Fig. 2. Game screenshot. The player is aiming at the enemy

The experience has three phases. The first one is a questionnaire about the player’s gaming habits and self-efficacy profile. This questionnaire has been used in an experimentation on Players’ confidence [7], but we removed the last part on risk aversion. It is to note that this questionnaire is in French, as our participants are all French-speaking people. The second part is the game session. Participants will have to play for at least 10 min and up to 20 min. At the start of the session, we pick a random difficulty curve between our four curves. During the game session, the player will face an AI-controlled enemy in an arena fight. When 5 min pass, a button appears on the top right of the screen, asking if the player is bored. If the player clicks on it, the game will select another difficulty curve. After 5 more minutes from the click, the button will reappear. Clicking again on the button will end the game session. We brief the players before the game session about the button, so they cannot miss it while playing. By doing this, we want to register a playtime, and consider that the playtime is a way to indirectly evaluate the players’ motivation.

Difficulty Pacing Impact on Player Motivation

147

No matter how the game session ended, by clicking the buttons or by playing 20 min, the player will enter the third phase of the test, and fill is the Game Experience Questionnaire or GEQ [15] that has been translated into French. The experimentation sessions took place at LeCNAM ENJMIN, a video game school [10]. Most students there play video games and use 3D software, so our participants already know how to navigate a 3D world and play a FPS with a PC setup. Indeed, learning these skills on a 20 min game session would not be representative of a real play session.

5

Results

Sixty-seven participants played our shooter game (55 male and 12 female) with a mean age of 23 (σ = 3.24). Participants played for at least 10 min, and up to 20 min, with a mean playtime of 17 min and 40 s. We estimate the DDA model’s quality for each participant. To do so, we calculate the ratio of the number of prediction errors to the total number of predictions for each play session. We decided to remove extreme values, using the interquartile range (IQR) of the statistical dispersion of the model’s quality1 We removed 8 participants with a too high prediction error ratio. We checked the players’ performance by calculating the mean difficulty of their tries, in order to detect if some players did not understand the rules of the game, or on the contrary, if they found an exploit that would lead them to victory at each try. But we did not find any abnormal data. We have three hypotheses to test, using the data we collected. Our first hypothesis concerns the baseline difficulty of the curves. We are thinking that participants will enjoy the game more with a higher difficulty level, that will put them in the flow state or arousal state. Those two states should motivate the players or at least make them play longer, which means that we expect participants to play longer on the Flow curve and Flow Peaks curve. Our second hypothesis, following the literature, is that participants played longer on curves with difficulty peaks. We also expect participants to play longer on high difficulty curves, and on difficulty curves with peaks. Combining these two statements, we expect the Flow Peaks curve to be the more appealing curve of all four. In order to check our hypotheses, we checked for each participant how long they played on the first curve, before clicking on the button for the first time. Because the player could click on a button to tell us when they are getting bored, we use the playtime as an estimation of the player enjoyment of the game for this curve. As shown by Fig. 3, most of the participants who had the Flow curve played less, meaning they pressed the button earlier, while participants with Low, Low Peaks and Flow Peaks curves played longer before pressing the button. 1

To find the extreme value, we get the upper extremity of the dispersion: Quartile 3 + 1,5*IQR.

148

W. R. Fernandes and G. Levieux

Fig. 3. Playtime of the participants on the first curve chosen Table 1. P-value of the Wilcoxon Rank Sum tests comparing the playtime for each couples of curves. The Flow Curve is significantly lower than the two curves with peaks, no matter the overall difficulty. * for p